| | |
| | |  |
| | | |
| | | # Yolo-Windows v2 |
| | | # "You Only Look Once: Unified, Real-Time Object Detection (version 2)" |
| | | A yolo windows version (for object detection) |
| | | # Magic: The Gathering Card Detection Model |
| | | |
| | | Contributtors: https://github.com/pjreddie/darknet/graphs/contributors |
| | | This is a fork of [Yolo-v3 and Yolo-v2 for Windows and Linux by AlexeyAB](https://github.com/AlexeyAB/darknet#how-to-compile-on-linux) for creating a custom model for [My MTG card detection project](https://github.com/hj3yoo/MTGCardDetector). |
| | | |
| | | This repository is forked from Linux-version: https://github.com/pjreddie/darknet |
| | | ## Day ~0: Sep 6th, 2018 |
| | | |
| | | More details: http://pjreddie.com/darknet/yolo/ |
| | | Uploading all the progresses on the model training for the last few days. |
| | | |
| | | ##### Requires: |
| | | * **MS Visual Studio 2015 (v140)**: https://www.microsoft.com/download/details.aspx?id=48146 |
| | | * **CUDA 8.0 for Windows x64**: https://developer.nvidia.com/cuda-downloads |
| | | * **OpenCV 2.4.9**: https://sourceforge.net/projects/opencvlibrary/files/opencv-win/2.4.9/opencv-2.4.9.exe/download |
| | | - To compile without OpenCV - remove define OPENCV from: Visual Studio->Project->Properties->C/C++->Preprocessor |
| | | - To compile with different OpenCV version - change in file yolo.c each string look like **#pragma comment(lib, "opencv_core249.lib")** from 249 to required version. |
| | | - With OpenCV will show image or video detection in window and store result to: test_dnn_out.avi |
| | | First batch of model training is completed, where I used ~40,000 generated images of MTG cards laid out in one of the pre-defined pattern. |
| | | |
| | | ##### Pre-trained models for different cfg-files can be downloaded from (smaller -> faster & lower quality): |
| | | * `yolo.cfg` (256 MB COCO-model) - require 4 GB GPU-RAM: http://pjreddie.com/media/files/yolo.weights |
| | | * `yolo-voc.cfg` (256 MB VOC-model) - require 4 GB GPU-RAM: http://pjreddie.com/media/files/yolo-voc.weights |
| | | * `tiny-yolo.cfg` (60 MB COCO-model) - require 1 GB GPU-RAM: http://pjreddie.com/media/files/tiny-yolo.weights |
| | | * `tiny-yolo-voc.cfg` (60 MB VOC-model) - require 1 GB GPU-RAM: http://pjreddie.com/media/files/tiny-yolo-voc.weights |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_training_set_example_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_training_set_example_2.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_training_set_example_3.jpg" width="360"> |
| | | |
| | | Put it near compiled: darknet.exe |
| | | After 5000 training epochs, the model got 88% validation accuracy on the generated test set. |
| | | |
| | | You can get cfg-files by path: `darknet/cfg/` |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_2.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_3.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_4.jpg" width="360"> |
| | | |
| | | ##### Examples of results: |
| | | However, there are some blind spots on the model, notably: |
| | | |
| | | [](https://www.youtube.com/watch?v=VOC3huqHrss "Everything Is AWESOME") |
| | | - Fails to spot some of the obscured cards, where only a fraction of them are shown. |
| | | - Fairly fragile against any glaring or light variations. |
| | | - Cannot detect any skewed cards. |
| | | |
| | | Others: https://www.youtube.com/channel/UC7ev3hNVkx4DzZ3LO19oebg |
| | | Example of bad detections: |
| | | |
| | | ### How to use: |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_5.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_6.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/0_detection_result_7.jpg" width="360"> |
| | | |
| | | ##### Example of usage in cmd-files from `build\darknet\x64\`: |
| | | The second and third problems should easily be solved by further augmenting the dataset with random lighting and image skew. I'll have to think more about the first problem, though. |
| | | |
| | | * `darknet_voc.cmd` - initialization with 256 MB VOC-model yolo-voc.weights & yolo-voc.cfg and waiting for entering the name of the image file |
| | | * `darknet_demo_voc.cmd` - initialization with 256 MB VOC-model yolo-voc.weights & yolo-voc.cfg and play your video file which you must rename to: test.mp4, and store result to: test_dnn_out.avi |
| | | * `darknet_net_cam_voc.cmd` - initialization with 256 MB VOC-model, play video from network video-camera mjpeg-stream (also from you phone) and store result to: test_dnn_out.avi |
| | | * `darknet_web_cam_voc.cmd` - initialization with 256 MB VOC-model, play video from Web-Camera number #0 and store result to: test_dnn_out.avi |
| | | ## Sept 7th, 2018 |
| | | |
| | | ##### How to use on the command line: |
| | | * 256 MB COCO-model - image: `darknet.exe detector test data/coco.data yolo.cfg yolo.weights -i 0 -thresh 0.2` |
| | | * Alternative method 256 MB COCO-model - image: `darknet.exe detect yolo.cfg yolo.weights -i 0 -thresh 0.2` |
| | | * 256 MB VOC-model - image: `darknet.exe detector test data/voc.data yolo-voc.cfg yolo-voc.weights -i 0` |
| | | * 256 MB COCO-model - video: `darknet.exe detector demo data/coco.data yolo.cfg yolo.weights test.mp4 -i 0` |
| | | * 256 MB VOC-model - video: `darknet.exe detector demo data/voc.data yolo-voc.cfg yolo-voc.weights test.mp4 -i 0` |
| | | * Alternative method 256 MB VOC-model - video: `darknet.exe yolo demo yolo-voc.cfg yolo-voc.weights test.mp4 -i 0` |
| | | * 60 MB VOC-model for video: `darknet.exe detector demo data/voc.data tiny-yolo-voc.cfg tiny-yolo-voc.weights test.mp4 -i 0` |
| | | * 256 MB COCO-model for net-videocam - Smart WebCam: `darknet.exe detector demo data/coco.data yolo.cfg yolo.weights http://192.168.0.80:8080/video?dummy=param.mjpg -i 0` |
| | | * 256 MB VOC-model for net-videocam - Smart WebCam: `darknet.exe detector demo data/voc.data yolo-voc.cfg yolo-voc.weights http://192.168.0.80:8080/video?dummy=param.mjpg -i 0` |
| | | * 256 MB VOC-model - WebCamera #0: `darknet.exe detector demo data/voc.data yolo-voc.cfg yolo-voc.weights -c 0` |
| | | Added several image augmentation techniques to apply to the training set: noise, dropout, light variation, and glaring: |
| | | |
| | | ##### For using network video-camera mjpeg-stream with any Android smartphone: |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_augmented_set_example_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_augmented_set_example_2.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_augmented_set_example_3.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_augmented_set_example_4.jpg" width="360"> |
| | | |
| | | 1. Download for Android phone mjpeg-stream soft: IP Webcam / Smart WebCam |
| | | Currently trying to generate enough images to start model training. Hopefully this helps. |
| | | |
| | | Recompiled darknet with OpenCV and CUDNN installed, and recalculated anchors. |
| | | |
| | | Smart WebCam - preferably: https://play.google.com/store/apps/details?id=com.acontech.android.SmartWebCam |
| | | IP Webcam: https://play.google.com/store/apps/details?id=com.pas.webcam |
| | | ----------------------- |
| | | |
| | | 2. Connect your Android phone to computer by WiFi (through a WiFi-router) or USB |
| | | 3. Start Smart WebCam on your phone |
| | | 4. Replace the address below, on shown in the phone application (Smart WebCam) and launch: |
| | | I've ran a quick training with tiny_yolo configuration with new training data, and Voila! The model performs significantly better than the last iteration, even against some hard images with glaring & skew! The first prediction model can't detect anything from these new test images, so this is a huge improvement to the model :) |
| | | |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_detection_result_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_2.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_3.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_4.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_5.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_decision_result_6.jpg" width="360"> |
| | | |
| | | * 256 MB COCO-model: `darknet.exe detector demo data/coco.data yolo.cfg yolo.weights http://192.168.0.80:8080/video?dummy=param.mjpg -i 0` |
| | | * 256 MB VOC-model: `darknet.exe detector demo data/voc.data yolo-voc.cfg yolo-voc.weights http://192.168.0.80:8080/video?dummy=param.mjpg -i 0` |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/1_learning_curve.jpg" width="640"> |
| | | |
| | | The video demo can be found here: https://www.youtube.com/watch?v=kFE_k-mWo2A&feature=youtu.be |
| | | |
| | | ### How to compile: |
| | | |
| | | 1. If you have CUDA 8.0, OpenCV 2.4.9 (C:\opencv_2.4.9) and MSVS 2015 then start MSVS, open `build\darknet\darknet.sln` and do the: Build -> Build darknet |
| | | ## Sept 10th, 2018 |
| | | |
| | | 2. If you have other version of CUDA (not 8.0) then open `build\darknet\darknet.vcxproj` by using Notepad, find 2 places with "CUDA 8.0" and change it to your CUDA-version, then do step 1 |
| | | I've been training a new model with a full YOLOv3 configuration (previous one used Tiny YOLOv3), and it's been taking a lot more resources: |
| | | |
| | | 3. If you have other version of OpenCV 2.4.x (not 2.4.9) then you should change pathes after `\darknet.sln` is opened |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/2_learning_curve.jpg" width="640"> |
| | | |
| | | 3.1 (right click on project) -> properties -> C/C++ -> General -> Additional Include Directories |
| | | |
| | | 3.2 (right click on project) -> properties -> Linker -> General -> Additional Library Directories |
| | | The author of darknet did mention that full network will take significantly more training effort, so I'll just have to wait. At this rate, it should reach 50k epoch in about a week :/ |
| | | |
| | | 4. If you have other version of OpenCV 3.x (not 2.4.x) then you should change many places in code by yourself. |
| | | |
| | | ### How to compile (custom): |
| | | ## Sept 13th, 2018 |
| | | |
| | | Also, you can to create your own `darknet.sln` & `darknet.vcxproj`, this example for CUDA 8.0 and OpenCV 2.4.9 |
| | | The training for full YOLOv3 model has turned sour - the loss saturated around 0.45, and didn't seem like it would improve in any reasonable amount of time. |
| | | |
| | | Then add to your created project: |
| | | - (right click on project) -> properties -> C/C++ -> General -> Additional Include Directories, put here: |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/3_learning_curve.jpg" width="640"> |
| | | |
| | | `C:\opencv_2.4.9\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir);$(cudnn)\include` |
| | | - right click on project -> Build dependecies -> Build Customizations -> set check on CUDA 8.0 or what version you have - for example as here: http://devblogs.nvidia.com/parallelforall/wp-content/uploads/2015/01/VS2013-R-5.jpg |
| | | - add to project all .c & .cu files from `\src` |
| | | - (right click on project) -> properties -> Linker -> General -> Additional Library Directories, put here: |
| | | |
| | | `C:\opencv_2.4.9\opencv\build\x64\vc12\lib;$(CUDA_PATH)lib\$(PlatformName);$(cudnn)\lib\x64;%(AdditionalLibraryDirectories)` |
| | | - (right click on project) -> properties -> Linker -> Input -> Additional dependecies, put here: |
| | | |
| | | `..\..\3rdparty\lib\x64\pthreadVC2.lib;cublas.lib;curand.lib;cudart.lib;cudnn.lib;%(AdditionalDependencies)` |
| | | - (right click on project) -> properties -> C/C++ -> Preprocessor -> Preprocessor Definitions |
| | | |
| | | `OPENCV;_TIMESPEC_DEFINED;_CRT_SECURE_NO_WARNINGS;GPU;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)` |
| | | - compile to .exe (X64 & Release) and put .dll-s near with .exe: |
| | | |
| | | `pthreadVC2.dll, pthreadGC2.dll` from \3rdparty\dll\x64 |
| | | |
| | | `cusolver64_80.dll, curand64_80.dll, cudart64_80.dll, cublas64_80.dll` - 80 for CUDA 8.0 or your version, from C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin |
| | | |
| | | |
| | | ## How to train (Pascal VOC Data): |
| | | |
| | | 1. Download pre-trained weights for the convolutional layers (76 MB): http://pjreddie.com/media/files/darknet19_448.conv.23 and put to the directory `build\darknet\x64` |
| | | |
| | | 2. Download The Pascal VOC Data and unpack it to directory `build\darknet\x64\data\voc`: http://pjreddie.com/projects/pascal-voc-dataset-mirror/ will be created file `voc_label.py` and `\VOCdevkit\` dir |
| | | |
| | | 3. Download and install Python for Windows: https://www.python.org/ftp/python/3.5.2/python-3.5.2-amd64.exe |
| | | As expected, the performance of the model with 0.45 loss was fairly bad. Not to mention that it's quite slower, too. I've decided to continue with tiny YOLOv3 weights. I tried to train it further, but it was already saturated, and was the best it could get. |
| | | |
| | | 4. Run command: `python build\darknet\x64\data\voc\voc_label.py` (to generate files: 2007_test.txt, 2007_train.txt, 2007_val.txt, 2012_train.txt, 2012_val.txt) |
| | | --------------------- |
| | | |
| | | 5. Run command: `type 2007_train.txt 2007_val.txt 2012_*.txt > train.txt` |
| | | Bad news, I couldn't find any repo that has python wrapper for darknet to pursue this project further. There is a [python example](https://github.com/AlexeyAB/darknet/blob/master/darknet.py) in the original repo of this fork, but [it doesn't support video input](https://github.com/AlexeyAB/darknet/issues/955). Other darknet repos are in the same situation. |
| | | |
| | | 6. Start training by using `train_voc.cmd` or by using the command line: `darknet.exe detector train data/voc.data yolo-voc.cfg darknet19_448.conv.23` |
| | | I suppose there is a poor man's alternative - feed individual frames from the video into the detection script for image. I'll have to give it a shot. |
| | | |
| | | If required change pathes in the file `build\darknet\x64\data\voc.data` |
| | | |
| | | More information about training by the link: http://pjreddie.com/darknet/yolo/#train-voc |
| | | ## Sept 14th, 2018 |
| | | |
| | | ## How to train with multi-GPU: |
| | | Thankfully, OpenCV had an implementation for DNN, which supports YOLO as well. They have done quite an amazing job, and the speed isn't too bad, either. I can score about 20~25fps on my tiny YOLO, without using GPU. |
| | | |
| | | 1. Train it first on 1 GPU for like 1000 iterations: `darknet.exe detector train data/voc.data yolo-voc.cfg darknet19_448.conv.23` |
| | | |
| | | 2. Then stop and by using partially-trained model `/backup/yolo-voc_1000.weights` run training with multigpu (up to 4 GPUs): `darknet.exe detector train data/voc.data yolo-voc.cfg yolo-voc_1000.weights -gpus 0,1,2,3` |
| | | ## Sept 15th, 2018 |
| | | |
| | | https://groups.google.com/d/msg/darknet/NbJqonJBTSY/Te5PfIpuCAAJ |
| | | I tried to do an alternate approach - instead of making model identify cards as annonymous, train the model for EVERY single card. As you may imagine, this isn't sustainable for 10000+ different cards that exists in MTG, but I thought it would be reasonable for classifying 10 different cards. |
| | | |
| | | ## How to train (to detect your custom objects): |
| | | Result? Suprisingly effective. |
| | | |
| | | 1. Create file `yolo-obj.cfg` with the same content as in `yolo-voc.cfg` (or copy `yolo-voc.cfg` to `yolo-obj.cfg)` and: |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_detection_result_1.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_detection_result_2.jpg" width="360"><img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_detection_result_3.jpg" width="360"> <img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_detection_result_4.png" width="360"> |
| | | |
| | | * change line `classes=20` to your number of objects |
| | | * change line `filters=425` to `filters=(classes + 5)*5` (generally this depends on the `num` and `coords`, i.e. equal to `(classes + coords + 1)*num`) |
| | | They're of course slightly worse than annonymous detection and impractical for any large number of cardbase, but it was an interesting approach. |
| | | |
| | | For example, for 2 objects, your file `yolo-obj.cfg` should differ from `yolo-voc.cfg` in such lines: |
| | | ------------------ |
| | | |
| | | ``` |
| | | [convolutional] |
| | | filters=35 |
| | | I've made a quick openCV algorithm to extract cards from the image, and it works decently well: |
| | | |
| | | [region] |
| | | classes=2 |
| | | ``` |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/4_detection_result_5.jpg" width="360"> |
| | | |
| | | 2. Create file `obj.names` in the directory `build\darknet\x64\data\`, with objects names - each in new line |
| | | At the moment, it's fairly limited - the entire card must be shown without obstruction nor cropping, otherwise it won't detect at all. |
| | | |
| | | 3. Create file `obj.data` in the directory `build\darknet\x64\data\`, containing (where **classes = number of objects**): |
| | | Unfortunately, there is very little use case for my trained network in this algorithm. It's just using contour detection and perceptual hashing to match the card. |
| | | |
| | | ``` |
| | | classes= 2 |
| | | train = train.txt |
| | | valid = test.txt |
| | | names = obj.names |
| | | backup = backup/ |
| | | ``` |
| | | |
| | | 4. Put image-files (.jpg) of your objects in the directory `build\darknet\x64\data\obj\` |
| | | ## Sept 16th, 2018 |
| | | |
| | | 5. Create `.txt`-file for each `.jpg`-image-file - with the same name, but with `.txt`-extension, and put to file: object number and object coordinates on this image, for each object in new line: `<object-class> <x> <y> <width> <height>` |
| | | I've tweaked the openCV algorithm from yesterday and ran for a demo: |
| | | |
| | | Where: |
| | | * `<object-class>` - integer number of object from `0` to `(classes-1)` |
| | | * `<x> <y> <width> <height>` - float values relative to width and height of image, it can be equal from 0.0 to 1.0 |
| | | * atention: `<x> <y>` - are center of rectangle (are not top-left corner) |
| | | https://www.youtube.com/watch?v=BZkRZDyhMRE&feature=youtu.be |
| | | |
| | | For example for `img1.jpg` you should create `img1.txt` containing: |
| | | ## Oct 4th, 2018 |
| | | |
| | | ``` |
| | | 1 0.716797 0.395833 0.216406 0.147222 |
| | | 0 0.687109 0.379167 0.255469 0.158333 |
| | | 1 0.420312 0.395833 0.140625 0.166667 |
| | | ``` |
| | | With the current model I have, there seems to be little hope - I simply don't have enough knowledge in classical CV technique to separate overlaying cards. Even if I could, perceptual hash will be harder to use if I were to use only a fraction of a card image to classify it. |
| | | |
| | | 6. Create file `train.txt` in directory `build\darknet\x64\data\`, with filenames of your images, each filename in new line, with path relative to `darknet.exe`, for example containing: |
| | | There is an alternative to venture into instance segmentation with [mask R-CNN](https://arxiv.org/pdf/1703.06870.pdf), at the cost of losing real-time processing speed (and considerably more development time). Maybe worth a shot, although I'd have to nearly start from scratch (other than training data generation). |
| | | |
| | | ``` |
| | | data/obj/img1.jpg |
| | | data/obj/img2.jpg |
| | | data/obj/img3.jpg |
| | | ``` |
| | | ## Oct 10th, 2018 |
| | | |
| | | 7. Download pre-trained weights for the convolutional layers (76 MB): http://pjreddie.com/media/files/darknet19_448.conv.23 and put to the directory `build\darknet\x64` |
| | | I've been trying to fiddle with the mask R-CNN using [this repo](https://github.com/matterport/Mask_RCNN)'s implementation, and got to train them with 60 manually labelled image set. The result is not too bad considering such a small dataset was used. However, there was a high FP rate overall (again, probably because of small dataset and the simplistic features of cards). |
| | | |
| | | 8. Start training by using the command line: `darknet.exe detector train data/obj.data yolo-obj.cfg darknet19_448.conv.23` |
| | | <img src="https://github.com/hj3yoo/darknet/blob/master/figures/5_rcnn_result_1.jpg" width="360"><img src="https://github.com/hj3yoo/darknet/blob/master/figures/5_rcnn_result_2.jpg" width="360"><img src="https://github.com/hj3yoo/darknet/blob/master/figures/5_rcnn_result_3.jpg" width="360"><img src="https://github.com/hj3yoo/darknet/blob/master/figures/5_rcnn_result_4.jpg" width="360"><img src="https://github.com/hj3yoo/darknet/blob/master/figures/5_rcnn_result_5.jpg" width="360"> |
| | | |
| | | Although it may be worth to generate large training dataset and train the model more thoroughly, I'm being short on time, as there are other priorities to do. I may revisit this later. |