Download | StarNet
Download | StarNet
StarNet
StarNet official page
Menu
Skip to content
Home
Download
News
Resources
Tips & Tricks
Support
About me
Download
This page contains links to download official distributions of StarNet. See Resources to find helpful materials created by community. See Tips & Tricks for some advice on how to get best results using StarNet.
New: See new experimental CLI distribution for MacOS here.
PixInsight Plugin (v2.1.x)
From now on I am planning to have two kinds of downloads for PixInsight: update package that only contains newest StarNet-pxm library and no Tensorflow libraries. This should be used to replace previous version of StarNet installed without changing your existing installation. The fresh install package contains all the necessary libraries as before.
Older releases can be found here.
Windows
PI versions 1.8.9-1 and lower:
Update: StarNet2_Win_2.1.0_tf_x64_update.zipFresh install: StarNet2_Win_2.1.0_tf_x64_install.zip
PI versions 1.8.9-2 and higher:
Update: StarNet2_win_2.1.1-0126_TF_x64_update.zipFresh install: StarNet2_win_2.1.1-0126_TF_x64_install.zip
Mac OS
Notes for Mac OS installation:
If PixInsight does not find new module after you copied all the files, try copying libtensorflow.2.dylib and libtensorflow_framework.2.dylib into /Applications/PixInsight/PixInsight.app/Contents/Frameworks/If you use Finder to do this, you will need to navigate into /Applications/PixInsight/, right-click PixInsight and choose ‘Show Package Contents’ and go further from there as usual.There is a bug in PI for MacOS that results in an error installing process modules (not only StarNet). When installing using Process -> Modules -> Install Modules do not click on the StarNet module after you do search, just hit Search->Install. Otherwise, the installation will error out. If you did click on on module name to view its properties (confirm version, etc.) you will need to restart PI and try again.
PI versions 1.8.9-1 and lower:
Update: StarNet2_MacOS_2.1.0_tf_x64_update.zipFresh install: StarNet2_MacOS_2.1.0_tf_x64_install.zip
PI versions 1.8.9-2 and higher:
Please make sure to read the Note #2 above!
Update: StarNet2_MacOS_2.1.1-0126_TF_x64_updateFresh install: StarNet2_MacOS_2.1.1-0126_TF_x64_install.zip
Linux
PI versions 1.8.9-1 and lower:
Update: StarNet2_linux_2.1.0_tf_x64_update.zipFresh install: StarNet2_linux_2.1.0_tf_x64_install.zip
PI versions 1.8.9-2 and higher:
Update: StarNet2_linux_2.1.1-0126_TF_x64_update.zipFresh install: StarNet2_linux_2.1.1-0126_TF_x64_install.zip
Command Line Tool (v2.0.0)
Will be updated soon.
Windows
Fresh install: StarNetv2CLI_Win.zip
MacOS
Fresh install: StarNetv2CLI_MacOS.zip
Linux
Fresh install: StarNetv2CLI_linux.zip
Graphical User Interface
Windows
Fresh install: StarNetv2GUI_Win.zip
Proudly powered by WordPress
深空摄影后期利器:starnet++ - 知乎
深空摄影后期利器:starnet++ - 知乎切换模式写文章登录/注册深空摄影后期利器:starnet++vappu软件开发行业 从业人员首先废话不多说,先上前后对比图:这张照片拍摄的是冬季星空的巴纳德环,就是那个暗红色的半圆.照片的中心是猎户座(三个等间距连在一起的星星).使用的设备是sigma 50mm F1.4,a7r3(天文改机),架在赤道仪上完成.单张曝光30s,ISO2000,拍摄16张叠加. 具体怎么叠加,怎么调色可以参考其他人的文章,如果大家特别感兴趣星空后期,我后面也可以单独写一篇文章介绍一下.在做深空或者星野后期的时候,经常要反复做一件事件,就是缩星.因为现在相机的高感是在太好了,每次都能拍到太多密密麻麻的星星,而摄影师多数不是为了拍星星,而是拍星云(银河中性各种色彩,各种彩色星云,尘埃啥的).如果星星太多,反而会让星云不明显,有喧兵夺主的感觉.另外一个主要的原因,就是后期要强化星云的细节和色彩,星星太多,会严重影响最终效果.然后早几年就有各种photoshop的教程,教你怎么缩星,大多数都用到了PS里面的最小值滤镜:但这个功能很繁琐,还经常效果不好,前几年AI应用迅速发展,基于tensorflow的应用starnet++就出现了.它是一个开源免费的工具,可以在windows/Mac/linux上面运行,而且不需要外部工具的支持.这个工具的功能很简单:就是把星空照片中的星星都抹去,只留下星云,比如这个是官网的样片:Before:After:有了starless处理后的图片,就可以用各种工具,滤镜来处理星云,增加对比度,饱和度,锐化等等.最后如果觉得没有一颗星星实在太假,还是可以把原来的星星加回去,在PS里面叠加原来的图片,把混合模式设置成变亮就可以了.这里介绍一下这个小工具怎么使用.首先去下载对应OS的安装文件:Starnet++我这里介绍一下Linux版本(MacOS也是一样的,需要把rgb_starnet++变成可执行的权限就可以了)解压完可以看到一堆文件:$ ls
libtensorflow_framework.so.1 libtensorflow.so.1 libtiff.so.3 mono_starnet++ mono_test5.tif rgb_starnet++ rgb_test5.tif run_rgb_starnet.sh
libtensorflow_framework.so.1.14.0 libtensorflow.so.1.14.0 LICENSE mono_starnet_weights.pb README.txt rgb_starnet_weights.pb run_mono_starnet.sh starless.tif
有用的只有两个: rgb_starnet++, mono_starnet++rgb版本是用于处理彩色图片,mono用于处理黑白.在处理前,首先把图片输出成16bit tiff格式,这个在photoshop里面可以选:保存的时候选择非压缩格式,starnet不支持压缩的tiff.然后在命令行里把图片作为第一个参数:$ ./rgb_starnet++ ~/Downloads/DSC03041-DarkMedian-16bit2.tif
TIFFReadDirectory: Warning, /home/matrix/Downloads/DSC03041-DarkMedian-16bit2.tif: wrong data type 7 for "RichTIFFIPTC"; tag ignored.
Reading input image... Done!
Bits per sample: 16
Samples per pixel: 3
Height: 5304
Width: 7952
Restoring neural network checkpoint... Done!
2022-01-07 14:19:20.020030: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 AVX512F FMA
2022-01-07 14:19:20.048438: I tensorflow/core/platform/profile_utils/cpu_utils.cc:94] CPU Frequency: 2100000000 Hz
2022-01-07 14:19:20.053698: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x1c32470 executing computations on platform Host. Devices:
2022-01-07 14:19:20.053750: I tensorflow/compiler/xla/service/service.cc:175] StreamExecutor device (0):
Total number of tiles: 10375
2022-01-07 14:19:20.734289: W tensorflow/compiler/jit/mark_for_compilation_pass.cc:1412] (One-time warning): Not using XLA:CPU for cluster because envvar TF_XLA_FLAGS=--tf_xla_cpu_global_jit was not set. If you want XLA:CPU, either set that envvar, or use experimental_jit_scope to enable XLA:CPU. To confirm that XLA is active, pass --vmodule=xla_compilation_cache=1 (as a proper command-line flag, not via TF_XLA_FLAGS) or set the envvar XLA_FLAGS=--xla_hlo_profile.
100% finished
Done!整体的处理时间取决于你电脑的CPU能力,starnet最多会使用16个CPU处理一张图片(你有100个CPU也没用...剩下的84个核都在围观),在我的老服务器上,一张42M的a7r3的图片,需要处理5分钟. 因为我都是拷贝到服务器上处理,因此我的电脑还可以PS其他图片,完全不受影响.(在2019年的intel处理器的macbook pro上也差不多5分钟,M1还没有试过)如果的CPU足够强悍(比如我的服务器有89个核,我就是同时处理几个starnet,一杯咖啡过后,就可以看结果了)rgb_starnet++还有一些高级参数,比如STRIDE:mono_starnet++ INPUT OUTPUT STRIDE默认64,这个值越小,处理出来的图片精细度越高,也意味着处理时间越长.我尝试过设置成32,一张图片处理了20分钟还没100%,就放弃了.最后分享一点小技巧,就是最好把原图直接用starnet来消星,而不是深度后期过的图片.我试过几次用深度后期的图片来消星,图片会非常赃,有很多灰色的孔洞,手工用修补工具处理得不偿失.使用原图消星,过渡很自然具体使用可以参考README.txt.编辑于 2022-07-05 21:59摄影后期技术后期制作星空赞同 55 条评论分享喜欢收藏申请
福建星网锐捷通讯股份有限公司 | 融合创新科技，构建智慧未来
福建星网锐捷通讯股份有限公司 | 融合创新科技，构建智慧未来
首页
产品与方案
关于我们
加入我们
投资者关系
Ruijie |
Centerm |
NEXhome |
Akuvox |
Asiatelco |
Sinwo
智慧网络
极简网络
无线网络
IT管理
维护服务
极简以太全光网
智慧云
智慧金融
智慧云
支付
智慧通讯
统一通信
融合视讯
全光网络
智慧文娱
KTV娱乐
家庭娱乐
K米
魔欢音乐教育
智慧物联
智慧医疗
智慧工厂
智慧商贸
智慧城市
智慧校园
智慧营业厅
智慧社区
智能家居
智慧社区
智慧园区
智慧餐饮
AI视觉结算台
智慧食堂
云餐厅
智慧工厂
工业软件
AI工业硬件
仓储物流设备
星网锐捷(股票代码:002396)，秉承"融合创新科技，构建智慧未来"的经营理念，坚持以自主创新和行业应用为基础，是国内领先的ICT基础设施及AI应用方案提供商。
公司介绍
公司简介
研发体系
大 事 记
荣誉分享
管理方针
宣传资料平台
廉洁及合规举报
联系我们
服务热线
新闻中心
公司新闻
媒体报道
专题报道
信息公示
公司刊物
廉洁及合规举报
星网锐捷的舞台，始终属于任何勇于尝试，挑战自我的年轻人；他们的未来，就是星网锐捷的未来。
校园招聘
社会招聘
新闻与活动
星网锐捷闪耀“新加坡移动支付及电子银行展”
星网锐捷旗下升腾资讯携“智能POS、扫码POS、Linux POS、智能自助终端、云喇叭系列产品”以及丰富的数字化场景解决方案等创新成果强势亮相，向全球客户展示了代表国际支付行业发展趋势的前瞻产品，让现场来宾零距离体验创新科技。
创新为您，连接未来 | 星网锐捷亮相2024MWC
在西班牙巴塞罗那举行的2024年世界移动通信大会（MWC）上，星网锐捷旗下锐捷网络展示了将技术与应用充分融合的云数据中心、5G、光网络等产品及解决方案。
新 闻
创新为您，连接未来 | 星网锐捷亮相2024MWC
2024-02-28
Seamless Asia 2024 | 星网锐捷闪耀“新加坡移动支付及电子银行展”！
2024-02-23
星网锐捷连续3年荣获中国移动终端公司“明星产品奖”
2024-02-04
2024旗开得胜！升腾资讯斩获多项行业奖项！
2024-01-29
星网元智入选《福州市产业数字化服务商重点培育名单》！
2024-01-17
更多新闻
产品与方案推荐
极简以太全光解决方案3.0
极简光3.0采用光纤入室的部署方式，同时承袭2.X方案的以太彩光技术，能够支撑未来十年业务演进，保障应用体验和投资价值。
升腾智慧营业厅
升腾以创新技术形成智慧营业厅解决方案2.0，融合人工智能、生物识别、可信计算、透明计算、可视化大数据分析、云计算等高新技术。
K米，乐享欢聚
K米，做最好玩的k歌聚会娱乐平台，连接用户、商家.第三方服务商，向线下KTv输出“管理、流量、内容、广告”核心价值,构建互联网+聚会娱乐新生态。
NEXhome智能家居解决方案
NEXhome利用先进的网络通讯技术和智能控制技术，将与家居生活相关的各个子系统有机结合在一起，用户通过智能面板、手机APP、语音等方式就可以轻松管理和控制自己的家。
电子行业智能制造方案
星网元智“星智造”’电子行业解 决方案为面向电子产品生产全流程而设计，能够为用户提供“软件+硬件+服务”的三位一体的整体智能制造解决方案。
行业AIoT解决方案
星网物联面向金融、保险、教育、医疗等行业，提供融合互联网、多媒体、AI、物联网等新兴技术驱动的创新业务解决方案，为行业的数字转型和成长赋能。
关于星网锐捷
公司简介
研发体系
大 事 记
荣誉分享
管理方针
宣传资料平台
廉洁及合规举报
联系我们
服务热线
新闻中心
公司新闻
媒体报道
专题报道
信息公示
投资者关系
临时公告
定期公告
公司治理
内幕交易防控专栏
12-4法制宣传日专题
投资者保护宣传
独立董事工作动态
投资者联系
子站点
升腾资讯
锐捷网络
星网视易
星网智慧
星网物联
星网元智
星网天合
星微门
K米网
德明通讯
睿云联
关注我们
微信公众号
企业文化号
福建星网锐捷通讯股份有限公司 ©版权所有 2002-2024 站长信箱：manager@star-net.cn
闽ICP备10011049号
闽公网安备35010402350979
How to Download and Use StarNet++ for Astrophotography
How to Download and Use StarNet++ for Astrophotography
Skip to content
Blog
EquipmentExpand
Star Trackers
ZWO AM5
Sky-Watcher EQ6-R Pro
William Optics GT81 WIFD
Sky-Watcher Esprit 100
ZWO ASI2600MM Pro
Filters
Get StartedExpand
Deep-Sky Astrophotography How-To
The Best Telescope For Beginners
Astrophotography Telescopes
Choosing an Astrophotography Camera
Build a Deep-Sky Setup
My Top Posts
ResourcesExpand
Types of Telescopes
How Much Does a Telescope Cost?
Astrophotography Tips
Best Astronomy Apps
Astrophotography Books
Start Autoguiding
Practice Image Data
TutorialsExpand
Polar Alignment
How to Focus at Night
Deep Sky Stacker
Autoguiding
Image Processing TutorialExpand
Selective Color Boosting
Star Removal in Photoshop
HaRGB Processing
Narrowband Imaging
Milky Way Photography
Nightscape Photography
Image Stacking
Photographing the Moon
Search
Get the Guide
Toggle Menu
Use StarNet++ To Make Starless Astrophotography Images
StarNet++ is a simple program that allows you to remove the stars from your astrophotography images. I think it is an invaluable tool to have when processing your images, and I personally use it all the time.
I first heard about StarNet++ from Chuck Ayoub on his YouTube Channel. He used it to remove the stars from his image of the Wizard Nebula, and it worked astonishingly well.
Following that, I saw many discussions about StarNet++ and how to use it. Suddenly, I was seeing starless astrophotography images pop up on Instagram and Astrobin daily.
At first, I thought that the StarNet++ application was a plugin for PixInsight, but I soon realized that it was a stand-alone software that can be used without any additional software. What you choose to do with the starless image you create is up to you.
A starless version of the Tadpoles Nebula.
Download StarNet++ (Version 2)
Related: My Complete Astrophotography Image Processing Guide
Why It’s So Useful
The process of reducing or removing stars in an astrophoto is not a new concept in the astrophotography image processing world. This technique is used by many amateur astrophotographers to help reveal the delicate structures of their deep-sky object.
I even wrote a tutorial on removing stars in Photoshop, and it was my goto method for removing/reducing star sizes for many years. Now, I include StarNet++ as a part of my workflow at this stage.
By separating the stars from the nebula or galaxy in your images, you can control and manipulate the data in a more deliberate way overall. A quick example of this would be to brighten the outer regions of the Andromeda Galaxy without making the surrounding stars brighter and overpowering.
Certain subjects are better suited for a starless variation than others. For example, a dim nebula like NGC 7822 (at the top of this post) with small stars within the field.
Comparing my original image of the Orion Nebula with a starless version.
Subjects that include large, bright stars (such as the Orion Nebula shown above), can be tricky to process because the “glow” from the bright stars may be left behind.
Personally, I prefer to blend the starless version of my astrophotography image with a regular (“with stars”) version for a hybrid image. This can be a great way to add depth and impactful details to your subject.
In this StarNet++ tutorial, I will show you how to use the software, and how you can use the starless versions of your images to improve your photos.
View All Astrophotography Tutorials
Blending Narrowband Images with Color
StarNet++ is a useful way to blend narrowband images into your regular color (broadband) images. A traditional narrowband image will, of course, include the many stars that surround your deep-sky object.
By removing the stars from your narrowband image (such as h-alpha), you can avoid discolored stars by applying this layer without the worry of introducing black spots or a slight blur.
Combining narrowband images with a full-color image is something I regularly do to add an impactful punch to my images, and StarNet++ allows me to do this in a more efficient way.
Here is a look at an image captured in h-alpha with the stars removed. I can now apply this layer as a luminance layer to my color image to add a dramatic punch to the original image.
The Lagoon Nebula in H-Alpha (Stars removed).
Related: HaRGB Image Processing Tutorial.
How StarNet++ Works
The author (Nikita Misiura) has provided documentation for the software in this Cloudy Nights thread. Here, you’ll also see a number of example images and best practices for the tool.
Nikita’s description of this program is as follows:
“StarNet is a neural network that can remove stars from images in one simple step leaving only the background. More technically, it is a convolutional residual net with encoder-decoder architecture and with L1, Adversarial and Perceptual losses.”
There is also a PixInsight module for StarNet++, for users of that software. For Photoshop users like me, you can just use the application on its own, and simply drag-and-drop your images into the tool to remove the stars.
How to Remove Stars in Your Images
Running the StarNet++ tool on your images is very simple. You’ll first need to make sure that you have downloaded the tool (here), and extracted the files onto your computer.
In the image below, you’ll see all of the files the folder contains (along with a couple of my own images). The “rgb-starnet++” application file is what we are interested in (circled).
The software is very simple to install and run on a PC.
In order for StarNet++ to work, you’ll need your image to be in a specific file format. The image must be a 16-Bit TIF file. (I think a few other file types work, but stick with this exact format for now.)
You can confirm that your image is this file type and bit-depth in Photoshop under the “Image” menu. (Image > Mode > 16 Bits/Channel)
If your image is currently in 8 Bits/Channel in JPEG format, you can convert the image to 16 Bits/Channel and save it as a TIF in Photoshop. First, convert the images to 16 Bits/Channel (if necessary), then save it in the TIF file format.
Make sure your image is a 16 Bit TIF File before running StarNet++.
Once your image is in this format, you can move on to the next step.
All you need to do now is place the astrophotography image you want to run StarNet++ on in the directory where the application is stored. Then, drag and drop the image file onto the application icon (rgb_starnet++).
You will see a black window appear, and StarNet++ will begin the process of removing stars in your image. You can check the status bar to see the progress of your image.
Once the process has completed, look for a new file called “starless” in the directory. This is a copy of your original image, but without the stars!
If you have previously created a starless image using StarNet++, running the application again will overwrite your original “starless.tif” image.
The starless version of your image will populate in the StarNet directory.
Recommended Workflow
I find it useful to use the StarNet++ star removal tool at the end of my image processing workflow in the “finishing touches” stage. Here is my proposed workflow when using StarNet++ at this stage of the process.
Register and stack your data
Process your image with stars intact
Run the image through StarNet++
Enhance the deep-sky object in the starless image
Paste the starless image on top of the original as a new layer
Select the stars in your original image using a mask
Carefully blend the starless version with the original
Another method that was brought to my attention on Twitter, is to run the mono version of StarNet++ on the greyscale color channels of your image separately. Once you have run the software on all RGB channels, you can build the full-color image without the stars.
In my experience, this technique produces a better overall result because the tool can better isolate the stars in each color channel. It does take a little longer, however.
Here is an example of the results achieved using this method on the Lagoon Nebula. As I’ll discuss a little later, the starless grayscale channel images are a useful way to increase the brightness/color of specific areas of your image via a luminance layer.
Achieve better results on color images by running the RGB channels through StarNet++ separately.
You may also find it useful to create a starless version of your image during the star minimizing stage of your workflow. Minimizing stars is one of the last edits I make in Photoshop. Using StarNet++ to help create a version of your image with smaller stars is a practical way to perform this task.
For PixInsight Users:
Although I do not currently use PixInsight to process my images, regular contributor Brent Newton had this to say about using StarNet++:
“for the PixInsight master race, check the Sourceforge page for Starnet, it can be installed as a PixInsight process. It still only works on non-linear images but it’s a great way to bootstrap a good star mask. Said star mask with a little tweaking can then be used with Morph. Transformation and MLT to erode stars away in your linear images.”
Blending Starless Images into Your Original
Blending the two images can be tricky, and the process involves using star masks to do it properly (the Select and Mask tool is perfect for this). This can be a very time-consuming process, especially when there are bright areas of your subject (such as the core of the Orion Nebula) that glow at the same brightness as the stars.
Another challenge is dealing with the halos surrounding bright stars in the field. I recommend copying and pasting your original image as a layer on top of the starless version (as shown below).
Separate your starless version into a new layer in Photoshop.
You can simply dial back the opacity of the “with stars” layer if you want, and this will indeed reduce the impact of the bright stars in the field.
However, it will dim all of the stars in a very unnatural way. It’s not a great look. We need to make a precise selection around the stars to control their brightness and overall impact to the image.
To do this, I recommend creating a star mask using the color range tool in Photoshop. Select > Color Range > Highlights. Then, use the Select and Mask tool to refine your star mask as shown in the example below.
The original star mask that you created by selecting the highlights of the image will likely include the brightest areas of your nebula or galaxy. You’ll need to adjust the aggressiveness of the selection using the Fuzziness and Range sliders in the Color Range window.
A simple star mask in Photoshop using the “Select and Mask” tool.
For complete control over your image, create a star mask that allows you to copy just the stars in the image to a new layer. Then, you are able to adjust both the brightness and size of the stars independently from the rest of the image.
At this stage, try lowering the opacity of the original “with stars” image. Because you have preserved the brightness of the “important” stars in the image on the top layer, dimming the smaller stars creates a much more natural-looking image.
Your layers window should look similar to the one below.
My layers window in Adobe Photoshop 2020.
Of course, this technique can have varying degrees of success, depending on the type of object you use it on, and the precision of the star mask you create.
I believe the most powerful aspect of StarNet++ is how easy it is to separate the signal of your deep-sky object from the image. It allows you to manipulate the color and detail of your subject without worrying about increasing the size and brightness of the stars in the field.
I have found it to be a fantastic way to create a luminance layer. You can easily adjust the brightness of the highlights and mid-tones in your image without introducing star bloat or blown highlights.
Final Thoughts
Maybe you’ll use StarNet++ as a way to separate key elements of your astrophoto from each other for further processing. Maybe you enjoy the look of a completely starless image on its own.
Whatever your tastes lie, I think you’ll find StarNet++ to be a brilliant, lightweight, and free tool that you can add to your bag of tricks.
If you want to learn how I process each and every one of my astrophotography images from start to finish, please consider downloading my in-depth image processing guide.
A starless version of the Omega Nebula.
Helpful Resources:
Guide: Which Astrophotography Image Processing Software is Best?
Tutorial: How to Remove Gradients in Your Image.
Tutorial: How to Remove Stars in Photoshop (Manual Method).
Demonstration: Chuck’s video demonstrating StarNet++ on the Wizard Nebula:
Photo Gallery
About
Contact
Privacy Policy
© 2024 AstroBackyard
YouTube
Blog
EquipmentToggle child menuExpand
Star Trackers
ZWO AM5
Sky-Watcher EQ6-R Pro
William Optics GT81 WIFD
Sky-Watcher Esprit 100
ZWO ASI2600MM Pro
Filters
Get StartedToggle child menuExpand
Deep-Sky Astrophotography How-To
The Best Telescope For Beginners
Astrophotography Telescopes
Choosing an Astrophotography Camera
Build a Deep-Sky Setup
My Top Posts
ResourcesToggle child menuExpand
Types of Telescopes
How Much Does a Telescope Cost?
Astrophotography Tips
Best Astronomy Apps
Astrophotography Books
Start Autoguiding
Practice Image Data
TutorialsToggle child menuExpand
Polar Alignment
How to Focus at Night
Deep Sky Stacker
Autoguiding
Image Processing TutorialToggle child menuExpand
Selective Color Boosting
Star Removal in Photoshop
HaRGB Processing
Narrowband Imaging
Milky Way Photography
Nightscape Photography
Image Stacking
Photographing the Moon
Search
Toggle Menu Close
Search for:
Search
最新研究|IBM的StarNet将可解释的AI引入图像分类 - 知乎
最新研究|IBM的StarNet将可解释的AI引入图像分类 - 知乎切换模式写文章登录/注册最新研究|IBM的StarNet将可解释的AI引入图像分类扬奇智能社区作者：KYLE WIGGERS，Venturebeat， MARCH 19, 2020编译：Florence Wong， AICUG研究论文下载：关注公众号，回复“323”即可下载IBM的研究人员最近在http://Arxiv.org上发表的一篇论文中，提出了StarNet-一种端到端的可训练图像分类器。该分类器能够对它认为是支持其预测的关键区域进行定位。 除了解决视觉分类的任务外，StarNet还支持弱监督小样本目标检测任务，因此仅需要少量的嘈杂数据即可达到合理的精度。StarNet可以提高透明度，并减少新视觉领域（如自动驾驶汽车和自动工业机器人）所需的培训数据量。 通过扩展，它可以减少涉及分类器的AI项目的部署时间，调查显示时间范围在8到90天之间。StarNet包含一个连接到提取器的小样本分类器模块，这两个模块均以元学习的方式进行训练，其中从分类中随机抽取情节。 每个情节包括针对给定图像基本类（例如“乌龟”，“鹦鹉”，“鸡”和“狗”）的支持样本和随机查询样本。图说：StarNet通过在查询和小样本任务的支持图像之间找到较大的（明显地）匹配区域，从而为其预测提供证据，因此即使对于模棱两可的情况也可以提供合理的解释。 每个查询（顶部）和支持（底部）对的匹配区域均绘制为热图。StarNet尝试对每对支持和查询图像进行几何匹配，任意形状的区域匹配从两个图像到他们的局部变形（以适应形状变化）。 训练驱动匹配区域，使其与共享相同类别标签的图像对上存在的类别实例的位置相对应，从而对实例进行本地化。 在本地化后，StarNet会突出显示共同的图像区域，从而深入了解其如何做出预测。在实验中，研究人员仅使用类别标签进行训练，验证和所有支持图像，这些数据来自包括miniImageNet数据集，CIFAR-FS和FC100在内的数据集，他们均具有100种随机选择的类别。 CUB拥有11788张200种鸟类的图像； ImageNetLOC-FS，其中包括331种动物。他们在单一一张Nvidia K40图形卡上进行了2,000集验证和1000次测试，因此平均运行时间从每批1.15秒到每批2.2秒。在小样本的分类任务中，StarNet的性能要比最新基准高出5％。对于弱监督小样本的目标检测，该模型所获得的结果比所有比较基线所获得的结果有“大幅度提高”。该团队将这种强大的性能归功于StarNet通过本地化对对象进行分类的本领“未来的工作方向包括将StarNet扩展到有效的端到端可微分多尺度处理，以更好地处理非常小和非常大的物体。利用StarNet在培训过程中做出的位置预测进行迭代优化；并将StarNet应用于需要精确定位，却仅有少量样本的其他应用程序中，例如视觉跟踪。”人们通常认为，随着AI系统复杂性的提高，它的解释性总是越来越差。但是研究人员已经开始用Facebook的Captum之类的库挑战这一概念，该库解释了决策的制定可通过神经网络使用深度学习框架PyTorch，IBM的AI Explainability 360工具包和Microsoft的InterpretML。就Google而言，它最近详细介绍了一个解释图像分类器如何进行预测的系统，而OpenAI详细介绍了一种可视化AI决策的技术。发布于 2020-03-23 08:55人工智能IBM科技新闻赞同添加评论分享喜欢收藏申请
StarNet - MicroSurvey Software
StarNet - MicroSurvey Software
Products
Solutions
Dealers
Training
Support
News
Demos
Buy now
Select Page
Need to Find a Dealer?
Dealer Map
Need Marketing Resources?
MicroSurvey Product Resource Page
Software
AMS
Upgrades
Crossgrades
CCP / Network / Training
Desktop Software
SurveyTools for BricsCAD
MicroSurvey CAD
MicroSurvey embeddedCAD and inCAD
Adjustment Software
STAR*NET
Data Collection Software
FieldGenius
FieldGenius For Android
Forensic Software
Leica Map360
Software Maintenance
Annual Maintenance Subscription
3rd Party
HxGN Content Program
What type of surveying?
Cadastral
Topography
Civil Engineer
What’s New?
Pricing
Resources
Least Squares Done Right.
E
Request a Demo
SOLUTIONS
How Can STAR*NET Help?
Least Squares software
World Renowned
MicroSurvey’s STAR*NET is the world’s most recognized least squares adjustment software. STAR*NET has earned a reputation for being simple to use, powerfully accurate and reliable as a provider of least squares adjustment solutions. STAR*NET’s results serve as the benchmark for the industry, and is legislated as the mandatory adjustment package in some jurisdictions when used in plan submissions! Users often tell us that when they decided to add least squares adjustment to their workflow, implementing STAR*NET was the next logical step. Let STAR*NET help you improve both the precision of your work and your confidence; by using our world class least squares adjustment software, you can trust that you’ve got it right!
Features Used
Adjusted Observations
E
Network Plot
E
Accurate, verifiable
Outputs Coordinates
All surveying workflows allow you to compute positions. Using STAR*NET allows you to compute positions, identify the reliability of these positions, and easily sift through the data to reassure yourself that the results are true and free of systematic errors. When processing your measurements, STAR*NET gives you dynamic tips that identify problems and help you learn the required workflow. STAR*NET guides you through assessing your network, helping to ensure that all accepted measurements have the required “goodness of fit”, while comprehensive listing and output options allow you to document your process. This enables you to demonstrate to your clients that the results meet their standards, and that your reference can be relied upon as the pillar of their operation. When you add STAR*NET to your workflow, you’ll never find yourself feeling like you’ve thrown numbers into a “black box”, blindly trusting that the results are accurate; you’ll know they are!
Features Used
Find Point
E
Inline Error Messages
E
Multiple Sensors, Multiple Brands
Best solution for combining sensor types
From RPLS forums right down to our direct support team, our customers have told us over and over that “STAR*NET does one thing, least squares, and it does it well.” Least squares adjustments are provided by many solutions, though typically as a part of a package that works with proprietary data from proprietary sensors. STAR*NET provides you with a least squares adjustment solution that works with all your data sources and requirements. There are importers and conversion utilities for data from all major manufacturers. STAR*NET works in 1D (Level), 2D and 3D adjustments and simplifies combining and exporting both ground and grid data. STAR*NET is a practical tool for combining data from total stations, levels, GNSS and gyroscopes. Input data can be given weights to reflect input from old and new instruments, 2D and 3D data and even control points that have long since gone missing. With these robust features, it is no wonder why so many companies have been growing STAR*NET networks for 20+ years, providing a big asset to their company.
Features Used
Traverse
E
Post-Process
E
FEATURE INDEX
The Features You Need...
Full Standard Features List
IT WORKS
STAR*NET is the easiest Least Squares Adjustment package to use and understand
It integrates data easily
Conventional total station measurements can be easily combined with GPS vectors and leveling data
Input files are simple to work with and understand
The STAR*NET data input file can be created by many software packages. Records are one per line and can be edited with any text editor
Results are clear and reliable
STAR*NET results are well formatted for quick review.
Confidence in your work
Geo-referencing your surveys is much easier than working locally. Reports output both Geodetic and Ground coordinates with combined scale factors
Data converters for all types of formats
Designed for converting raw data from various data collection packages into STAR*NET data format
Contact Us
1-800-668-3312
Quick LinksPrivacy Policy
Terms of Services
End Agreement
Demo Install
Cookie Policy
Portal
About UsCareers
Company
News
Find a Dealer
FollowFollowFollow
Copyright © 2024 | MicroSurvey Software Inc. - part of Hexagon
Just a moment...
a moment...Enable JavaScript and cookies to contiSiril - Starnet++ integration
Siril - Starnet++ integration
Siril
About
(Current)
News
(Current)
Docs
(Current)
FAQ
(Current)
Tutorials
(Current)
Wiki
(Current)
Forum
(Current)
Downloads
(Current)
Shop
(Current)
� Search
(Current)
Donate
(Current)
Languages
Translations
English
Français
Starnet++ integration
By Adrian Knagg-Baugh
This tutorial aims to showcase use of the new Starnet++ integration available since Siril 1.2.0.
This tool meets the needs of users to be able to manipulate images from Starnet++ with Siril, including Pixel Math, the Star Recombination Tool and the Synthetic Star Tools.
New Siril Integration
Starnet++ Process
When to Remove Stars
Starnet++ User Interface and Options
The output files
Star Recombination
What about PixelMath?
My stars are terrible, can I save my image?
New Siril Integration #
In versions of Siril prior to 1.2.0 Starnet++ functionality was not integrated. A python script was available and covered here in a previous tutorial (now deleted), but it involved multiple additional dependencies, whereas the new integrated interface in 1.2.0 has no additional dependencies beyond requiring that Siril is compiled with libtiff support (this is the default, and if you are using packages your distribution will almost certainly have built it this way) and a working installation of Starnet++. Both v1 and v2 are supported.
Starnet++ Process #
Images from Siril pre-processing are usually in standard FITS/32bit format (though in some circumstances they may also be in FITS/16bit format).
However, the images usable by Starnet++ must be in TIFF/16bit format. They must therefore be converted.
Similarly, Starnet++ output images are in TIFF/16bit. If you want to benefit from Siril’s calculation precision, you have to convert them to FITS/32bit.
This format is also the only one supported by Siril’s PixelMath tool.
The new Starnet++ interface handles all the necessary image format conversion automatically. Unless the force 16 bit preference is set it will convert the output to 32 bit in order to be ready for increased precision calculations in subsequent processing.
The output image .fits FILTER headers are set to Starless and StarMask for the respective output files.
When to Remove Stars #
Starnet++ greatly prefers its input to have a midtone transfer function (MTF) autostretch applied - I speculate that this is how the data it was trained on was prepared. For this reason it is strongly recommended that you use Starnet++ before doing any custom stretching, and check the “Pre-stretch Linear Image” option.
There is an exception detailed on the Starnet++ website
: if you find difficulty with Starnet removing parts of nebulae along with the stars, you may have better results by maually applying a different MTF (histogram) stretch and turning off the pre-stretch option in the Starnet++ dialog. Note that in this case, if you have further linear processing to do you will need to note the values of the lo, mid and hi MTF sliders and apply the inverse stretch manually using the Siril console command invmtf.
The Starnet++ website also states that if dealing with colour images, it is better to remove stars from the full colour image rather than splitting the image into channels and removing stars from each in turn. It is therefore recommended that if processing mono images, they are composited into RGB colour before running Starnet++.
Starnet++ User Interface and Options #
The Starnet++ GUI is shown below. Note that the stride input control is only shown when Use custom stride is checked: it is unchecked by default.
Starnet++ interface GUI
Starnet++ can also be called from the console or scripts using the following command syntax:
starnet [-stretch]
[-upscale]
[-stride=value]
[-nostarmask]
The options work as follows:
-stride Starnet++ has an option to set the stride used to divide the image into chunks for processing. Reducing the stride greatly increases the processing time and the author of Starnet++ strongly recommends leaving it at the default. Nevertheless, the option is made available.
-upscale There is an option to upscale the input image by a factor of 2. This is provided following guidance on the Starnet++ website stating that upscaling may help if problems are experienced removing stars that cover a very small number of pixels. The option defaults to off.
The default option is to generate a star mask. This is simply the difference between the input image and the starless output, but it is extremely useful as part of a star reduction workflow where you wish to stretch the background separately to the stars but add the stars back later. If a star mask is not required the box may be unchecked in the GUI, or the option -nostarmask passed to the command.
Siril offers another tool, Star Recomposition, for simplifying the process of stretching the starless and star mask images independently. The Starnet++ GUI provides an option to launch this tool on completion of the Starnet++ process. Enabling this option will automatically enable star mask generation. There is no corresponding command option as this is purely a GUI tool.
Example of an image with both nebulous background and many stars
The output files #
On completion of the process the starless image is saved in the working directory with the same name as the original file, with the _starless preffix added. The starless image is loaded as the current working image in Siril. The star mask is saved in the current working directory with the _starmask preffix added to the original filename.
Example of the corresponding starless output file
Example of the corresponding star mask output file
Star Recombination #
Introduced in 1.2.0 alongside Starnet++ integration, the Star Recombination tool enables the blending of two separate images with each having different stretches applied. In theory it can be used to blend any two images together, but the primary design goal was to support workflows involving separate processing of the starless layer and the star mask.
The tool can be loaded in two ways: it is accessible through the Image Processing menu, and the Starnet++ dialog also has an option to launch it directly on completion of Starnet++.
The Star Recombination tool UI
The dialog takes the form of two columns, each with a set of generalised hyperbolic stretch (GHS) controls and a mini histogram display. The mini histograms also show a graph of the transfer function of each stretch, and may be minimized by clicking the drop arrow next to their title. The histograms can be toggled between linear and logarithmic mode and obey the global UI preference for default histogram mode. The recombination tool has no facility to stretch channels separately owing to space constraints in the UI. If this is desired, channels can still be stretched separately using the main GHS interface accessible through the Image Processing menu and then loading the stretched result into Star Recombination tool.
Applying stretch settings
As the full range of GHS controls is quite extensive, some have been hidden by default. They can be shown by clicking the “Advanced” button. The defaults have been selected so that excellent results can be achieved using the simple set of controls.
The controls behave the same as in the single-image Generalised Hyperbolic Stretch Transformations tool, with 2 exceptions. First, the black point control (available when Advanced is selected) applies to the base image as a separate stretch after the stretch set by the other controls. This means you will require a different BP value to achieve the same effect, but makes for a more natural workflow when combining the stretches on two different images. The other difference is that there is no ability to set the symmetry point SP by clicking on the histogram displays. This is because as a result of the different BP behaviour the meaning of the x axis in relation to pixel values in the original unstretched image is more ambiguous, so setting SP from it makes less sense.
Showing all settings
The two stretched images are combined in the CIE LAB color space using the following transformations (the subscripts l and r denote values from the left and right images respectively):
$L:L\rightarrow 1-((1-Ll)\times(1-Lr))$
$A:A\rightarrow \frac{Ll\times Al + Lr\times Ar}{Ll+Lr}$
$B:B\rightarrow \frac{Ll\times Bl + Lr\times Br}{Ll+Lr}$
What about PixelMath? #
The PixelMath
tool recently introduced in Siril makes it possible to recombine starless and star mask images in a different manner. The two files can be loaded into PixelMath and combined using the standard range of Siril’s PixelMath operations. Unlike the Star Recombination tool, PixelMath offers the option to be scripted using Siril’s command set. Full details of PixelMath can be found in the main PixelMath tutorial page.
My stars are terrible, can I save my image? #
Of course! Siril offers Synthetic star tools
which can help. If your stars are bad but for whatever reason it’s important to save the image rather than revisit the target, Synthetic Star tools can be used in conjunction with Starnet++ to replace them. Starnet++ is used to produce a starless image, and Synthetic star tools is used to generate a synthetic star mask using parameters of the stars actually measured from your image. The synthetic star mask can then be recombined with the starless image using the Star Recombination tool or PixelMath as described above.
It’s up to you now!!!
Follow us:
© Copyright 2024 Siril
PIXLS.US ♡ Siril
StarNet Star Removal — Siril 1.3.0 文档
StarNet Star Removal — Siril 1.3.0 文档
latest
通用
安装
Graphical User Interface
偏好设置
File Formats
Color Management
序列
（预）处理
Definitions and workflow
预处理
处理
Image stretching
Colors
Filters
Star Processing
StarNet Star Removal
指令
Star Recomposition
Desaturate Stars
完全再合成
Geometry
Background Extraction
Extraction
线性匹配
RGB compositing
Merge CFA Channels
Pixel Math
Scientific Analyses and tools
Plotting Feature
动态PSF
天体测量
测光
Intensity Profiling
Image Inspectors
统计信息
自动化
脚本
Headless mode
Path parsing
动态叠加
动态叠加
附录
Commands Reference
How to report issues
Siril
处理
Star Processing
StarNet Star Removal
在 GitLab 上编辑
StarNet Star Removal
StarNet is a software developed by Nikita Misiura. Its
first version was released under a free
and opensource license. Unfortunately, version 2 became proprietary and the
sources are since closed. The version 2 is available free of charge from
there. Make sure you download the
Command Line Tool version. Siril can interface with any version of the
StarNet CLI tool, including the new experimental Torch-based version that has
initially been released for M1- and M2- based Apple Macs.
警告
If you are wondering why StarNet doesn’t launch, please run it outside
Siril first. It’s not Siril’s fault if it’s not supported by your computer or
badly installed for some reason. If your processor does not support the
vectorization instructions required by StarNet, there is no way to bypass
that. The error message will be obtained by executing StarNet alone.
小技巧
On MacOS, for Siril to detect and use StarNet correctly, it is necessary
to fix some permissions and security issues first. Start by opening the
Terminal application from the Utilities folder within Applications. In
Terminal, you need to change your working directory from your home directory
to the StarNetCLI installation directory. To do this type in cd followed
by a space and then drag the StarNetCLI folder into the terminal
window to copy its path. Press enter. Then type in the following four
commands, pressing enter after each one:
xattr -r -d com.apple.quarantine libtensorflow_framework.2.dylib
xattr -r -d com.apple.quarantine starnet++
chmod +x starnet++
chmod +x run_starnet.sh
Then, at the first use with Siril, the execution of StarNet may fail with
a warning about libtensorflow. Cancel out of this warning. Open System
Preferences and under Privacy and Security click the Allow anyways
button for libtensorflow. After this, StarNet should execute properly in Siril.
小技巧
On MacOS, again, there is a new Starnet executable optimized for the Apple
Silicon chip that has been released on the site: https://www.starnetastro.com/experimental/.
This new version is much faster than previous version because it uses the
new MPS accelerated PyTorch (https://developer.apple.com/metal/pytorch/).
Also, this version contains signed binaries, follow the installation
instructions in the README.txt
However, it is still possible for Siril to run external binaries and this is
what we decided to implement starting with Siril 1.2.0. For the settings,
please refer to the preference page. It explains how to
tell Siril where StarNet is located.
警告
This is the location of the command line version of StarNet that need to
be given, not the GUI one.
Note that StarNet requires its input in the form of TIFF images, therefore if
Siril is compiled without libtiff support then the StarNet integration will
not be available.
The primary purpose of StarNet is to remove all the stars from the images in
order to apply a different process between the stars and the rest of the image.
This usually helps to control star bloat during the different stretches, but it
is also very useful for creating images of comets where the comet tracking
rate can be significantly different to the distant stars.
StarNet dialog box.
The tool is very easy to use and only five options are available:
Pre-stretch linear image: If selected, an optimised Midtone Transfer
Function (MTF) stretch is applied to the image before running StarNet, and
the inverse stretch is applied on completion. This is necessary for using
StarNet at the linear stage of processing.
Recompose stars on completion: If selected, on completion of the star
removal process the star recomposition tool will open, providing an interface
for independently stretching and blending the background and the stars if
star reduction, rather than total removal, is desired. This option has no
effect when processing a sequence.
Generate star mask: This will generate a star mask and saved it in the
working directory. The star mask is calculated as the difference between the
original image and the starless image. The default is to produce a star mask.
Upsample x2: This option will up-sample the image by a factor of 2
before running StarNet. This improves performance on very tight stars but
quadruples processing time and may impair performance on very large stars.
The image is rescaled to the original size on completion.
Use custom stride: A custom value may be entered for the StarNet
stride parameter. The default value is 256 and the StarNet developer
recommends not to change this.
The StarNet process can easily be applied to a sequence. The togglebutton
Apply to sequence selects whether the process will
apply to a single image or to a sequence. Where the process is applied to a
sequence, a new sequence will be created containing the starless images and, if
star mask generation is selected, a second sequence will be created containing
the corresponding star mask images.
More information about StarNet can be found on the original website.
A click on Execute will run the process. It can be slow, depending of
your machine performance. However, Siril shows a progressbar to follow the
processing. As with other Siril processes, if processing a sequence the progressbar
will only update after completion of each image in the sequence, and will show
the overall progress through the sequence.
指令
Siril command line
starnet [-stretch] [-upscale] [-stride=value] [-nostarmask]
Calls StarNet to remove stars from the loaded image.
Prerequisite: StarNet is an external program, with no affiliation with Siril, and must be installed correctly prior the first use of this command, with the path to its CLI version installation correctly set in Preferences / Miscellaneous.
The starless image is loaded on completion, and a star mask image is created in the working directory unless the optional parameter -nostarmask is provided.
Optionally, parameters may be passed to the command:
- The option -stretch is for use with linear images and will apply a pre-stretch before running StarNet and the inverse stretch to the generated starless and starmask images.
- To improve star removal on images with very tight stars, the parameter -upscale may be provided. This will upsample the image by a factor of 2 prior to StarNet processing and rescale it to the original size afterwards, at the expense of more processing time.
- The optional parameter -stride=value may be provided, however the author of StarNet strongly recommends that the default stride of 256 be used
Siril command line
seqstarnet sequencename [-stretch] [-upscale] [-stride=value] [-nostarmask]
This command calls Starnet++ to remove stars from the sequence sequencename. See STARNET
Links: starnet
上一页
下一页
© 版权所有 2023, Free-Astro Team。
版本 475510c3.
利用 Sphinx 构建，使用的
主题
由 Read the Docs 开发.
Read the Docs
v: latest
版本
latest
stable
下载
html
托管于 Read the Docs
项目主页
构建
How to use Starnet++ for PixInsight natively - Tutorial
How to use Starnet++ for PixInsight natively - Tutorial
top of pageLog InGALLERYOUR MESSIER CATALOGGALAXIESNEBULAESTAR CLUSTERSNIGHTTIME LANDSCAPESPLANETARY IMAGINGCONSTELLATIONSCOMETS & MORETUTORIALSALL TUTORIALSPREMIUM COURSEMILKY WAY PHOTOGRAPHYFULL ASTROPHOTOGRAPHY GUIDELEARN PIXINSIGHT2024 SOLAR ECLIPSESERVICESVIDEO CONSULTINGIMAGE PROCESSINGPRIVATE WORKSHOPREMOTE IMAGINGEQUIPMENTOUR EQUIPMENTBEGINNER EQUIPMENT GUIDEBEST BEGINNER TELESCOPESBEST BEGINNER CAMERASBEST MOUNTSGEAR REVIEWSVAONIS VESPERARESOURCESBLOGRAW DATANARROWBAND EXPECTATIONSBOOKSFREE TIPS EMAIL LISTFORUMABOUTABOUT USPRINTSAWARDSCHARLES MESSIER GRAVE PROJECTMoreUse tab to navigate through the menu items.V4 of our Beginner PixInsight Guide now available!All PostsMESSIER CATALOGNGC CATALOGIC CATALOGTUTORIALSLANDSCAPESPLANETARYGALAXIESNEBULAECLUSTERSFRONT PAGEARTICLESFRENCHREVIEWSBOOKSCONSTELLATIONSASTRONOMYSearchLog in / Sign upAntoine & Dalia GrelinAug 24, 20206 min readHow to use Starnet++ for PixInsight natively - TutorialUpdated: Jun 6, 2023How to install and use Starnet++ on PixInsight? In this post, I will go over the easiest way to get Starnet++ to work on PixInsight and give you a few tips on how to use it. You'll see that it is actually pretty easy.You can watch our video tutorial HERE.[Update 2022]Version 2 of Starnet is available. Be sure to read our updated guide on how to install StarNet V2 as the guide below is now outdated![Update 2023]We now recommend installing StarXTerminator instead, which is updated more often usually works better.Scroll to the bottom to see our full videoOn August 23rd, 2020, Pleiades Astrophoto released their newest update for PixInsight: Version 1.8.8-6! The main addition that caught my eye when reading through the list of new features is the fact that the popular star removal software, StarNet++ is now available in the standard version of PixInsight!Since its creation by Nikita Misiura in February 2019 (when Starnet++ was announced on a Cloudy Nights thread), the software was immediately loved by the amateur astrophotography community and quickly became popular. In the weeks that followed its announcement, you probably remember seeing dozens and dozens of new starless images popping up on your Astrobin feed, Instagram and other social media where you follow astrophotography content.Installing and running StarNet++ was never difficult, but it was pretty annoying especially for Mac users. I myself tried to run the program several times on my machine but ended up giving up just because it was too much trouble. Today, StarNet++ is available by default on PixInsight, but you still need to install a specific package to make it work, here are the steps to follow to get your first starless image! Install the latest version of PixInsightThe first thing you need to do is update PixInsight to the latest version (1.8.8-6 at the time of writing this tutorial). If you're a Mac user, first uninstall the application by dragging it from the Applications folder to the trash (make sure you do not trash the files in your documents folder though as those contain your license and user preferences).Next, head over to the PixInsight Software Distribution page and download the latest version for your computer. In our case, we downloaded the second option as we are using a Mac computer.After installing the package, launch PixInsight and see if any update window pops up as you may have to install a small update internally as well. Once this is done, close the program. Download the network weights for StarNetIn order to make StarNet++ work on PixInsight natively, you will just need to install two small .pb files on your computer. This is easily done by downloading the Starnet module folder for PixInsight which is available HERE.After downloading this folder, open it and locate the two .pb files:mono_starnet_weights.pbrgb_starnet_weights.pbThose are the ones we care about!Open a new window on your computer and locate your PixInsight folder, then open the Library folder inside it. On Windows, I assume it should be in your Program Files folder, but if you're on Mac, you need to go to:Applications > PixInsight > LibraryIn theory, the exact location shouldn't matter much but if you want to be safe, just follow our exact steps as it has been working fine for us Once you have the Library folder open, drag and drop the two files above into it. You can also place them in a "Starnet" subfolder if you wish to keep things clean and remember what those files are in the future.The location where we put our .pb files for Starnet on PixInsight Link the weights files to StarNet in PixInsightIf you are reading this, I feel like you're probably thinking "oh my, what now..?"Don't worry, this is the last and easiest step! Follow these steps:Open PixInsightGo to Process > StarNetClick on the little wrench icon For RGB weights file: Click on the folder icon and locate your rgb_starnet_weights.pb file (which should be in the library folder of PixInsight if you followed our steps earlier). Double click on it.For Grayscale weights file: Click on the folder icon and locate your mono_starnet_weights.pb file (which should be in the same folder as the RGB files). Double click on it.Click on OKYour StarNet process should now look like this:StarNet process with the correct weights files Get your first Starless image!The StarNet process is extremely easy to use. It does have one setting you can change called "Stride", but you should not have to and Nikita does not recommend playing with that number.When is the best time to use StarNet?All you need to do to execute StarNet is to drag it to an image (or click on the square icon). StarNet works best on Non-Linear images, meaning images that have already been fully stretched. It is also a good idea to use StarNet in the early stages of your processing workflow (as long as you are in the non linear phase). if you check to "Create Star Mask" box, you can use this star mask when you are done processing your starless data to add the stars back onto your image.To do that, simply open the PixelMath process and add your starless image and your star mask together using a "plus" sign. Then just apply the process!Example: "Starless+StarMask". Yep. Pretty basic.Below are four images. The first one shows the Pelican Nebula where StarNet was used at the very end (after all the processing was already done). The second one shows the Sadr Region, once again with StarNet being ran at the end.The third image shows the Heart Nebula with StarNet being applied right after switching to Non Linear. It is a bit difficult to see, but the Pelican Nebula and the Sadr Region still show quite a few stars whereas all the stars in the Heart Nebula image are gone. The last image is the Heart Nebula with the stars re-applied at the end of the processing workflow.The processing workflow used before and after running StarNet is the same one we always use, and can be found here).Below is a larger image of the Heart Nebula without stars, we love it!A Starless Heart Nebula Final ThoughtsStarNet is a great tool for astrophotography that can be used to remove stars in your images. Now that it is natively available for PixInsight, it has become extremely easy to use and is not only useful but very fun! Removing the stars in your images will allow you to boost the faint nebulosity in your object without blowing up any star. This will allow you to bring up detail much easier than before. I can't wait to reprocess some old data with the StarNet tool and see how much of a difference it makes on all types of targets!Make sure to follow us on Facebook, Instagram, Twitter and YouTube to stay up to date with our work!Clear Skies,Galactic HunterJoin our Patreon! GALACTIC HUNTER BOOKSThe Astrophotographer's Guidebook Description: Discover 60 Deep Sky Objects that will considerably improve your Imaging and Processing skills! Whether you are a beginner, intermediate, or advanced astrophotographer, this detailed book of the best deep sky objects will serve as a personal guide for years to come! Discover which star clusters, nebulae, and galaxies are the easiest and most impressive to photograph for each season. Learn how to find each object in the night sky, and read our recommendations on imaging them in a quick and comprehensive way. Each target in this guide contains our advice on imaging, photos of expected results, and a useful information table. We've also included a few cool facts about each target, a map to find it in the night sky, and more!
The Astrophotographer's JournalDescription: The Astrophotographer’s Journal is a portable notebook created for the purpose of recording observations, cataloguing photographs, and writing down the wonderful memories created by this hobby. This book contains more than 200 pages to memorialize your stargazing and imaging sessions, as well as a useful chart on the last pages to index exciting or important notes. Read back on the logs to see how much progress you have made through the months, the problems you overcame, and the notes taken to improve in the future. Just as the pioneers of astronomy did in their time, look up and take notes of your observations as you are the author of this star-filled journey.
The Constellations HandbookDescription: The Constellations Handbook is a logical guide to learning the 88 constellations. Learning the constellations is difficult. Remembering them is even harder. Have you ever wanted to look up to the night sky, name any pattern of stars and be able to tell their stories? This book groups the constellations in a logical order, so that the reader can easily learn them by their origin, and see how their stories interact with one another as a group. The last pages of this book include an index of all 88 constellations, each with a slot where you can write your own personal tips and tricks in order to memorize them with ease. The Constellations Handbook is not just another guide listing all the constellations from A to Z and their location, it is the perfect companion for stargazing, and a learning journey through the ages.
TUTORIALS•FRONT PAGE26,355 views0 comments1 like. Post not marked as liked1Related PostsSee AllHestia - Solar Observation Step-By-Step Guide5480Post not marked as likedDwarfLab Dwarf II - The Smallest and Most Affordable Smart Telescope - Full review79003 likes. Post not marked as liked3What is the Best Smart Telescope for Astrophotography2,72802 likes. Post not marked as liked2PROCESSING GUIDESGET OUR NEWSLETTERSubscribe to our newsletter and instantly receive 7 free wallpapers!Sign UpLATEST POSTSMessier 31 - The Andromeda Galaxy | Astrophotography and TipsHestia - Solar Observation Step-By-Step GuideM72 - Globular Cluster in Aquarius | Astrophotography Pictures and TipsTHE GALACTIC COURSELearn astrophotography with our premium online course. Get lifetime access to all types of imaging content!Disclosure: Some of the links on this website are affiliate links. This means that, at zero cost to you, we will earn an affiliate commission if you click through the link and finalize a purchase.Subscribe to our newsletter!Opt-in to get our quarterly newsletter and announcements.
After subscribing you'll receive an email with FREE wallpapers for your devices!EmailThanks! You are now subscribed!SIGN UP© 2016-2023 by Galactic Hunterbottom of page