Repository files navigation

YOLOv3 and YOLOv4 implementation in TensorFlow 2.x, with support for training, transfer training, object tracking mAP and so on...Code was tested with following specs:

• i7-7700k CPU and Nvidia 1080TI GPU
• OS Ubuntu 18.04
• CUDA 10.1
• cuDNN v7.6.5
• TensorRT-6.0.1.5
• Tensorflow-GPU 2.3.1
• Code was tested on Ubuntu and Windows 10 (TensorRT not supported officially)

First, clone or download this GitHub repository.Install requirements and download pretrained weights:

pip install -r ./requirements.txt# yolov3wget -P model_data https://pjreddie.com/media/files/yolov3.weights# yolov3-tinywget -P model_data https://pjreddie.com/media/files/yolov3-tiny.weights# yolov4wget -P model_data https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v3_optimal/yolov4.weights# yolov4-tinywget -P model_data https://github.com/AlexeyAB/darknet/releases/download/darknet_yolo_v4_pre/yolov4-tiny.weights

Start with using pretrained weights to test predictions on both image and video:

python detection_demo.py

mnist folder contains mnist images, create training data:

python mnist/make_data.py

./yolov3/configs.py file is already configured for mnist training.

Now, you can train it and then evaluate your model

python train.pytensorboard --logdir=log

Track training progress in Tensorboard and go tohttp://localhost:6006/:

Test detection withdetect_mnist.py script:

python detect_mnist.py

Results:

Custom training required to prepare dataset first, how to prepare dataset and train custom model you can read in following link:
https://pylessons.com/YOLOv3-TF2-custrom-train/
More about YOLOv4 training you can readon this link. I didn’t have time to implement all YOLOv4 Bag-Of-Freebies to improve the training process… Maybe later I’ll find time to do that, but now I leave it as it is. I recommended to useAlex's Darknet to train your custom model, if you need maximum performance, otherwise, you can use my implementation.

To learn more about Google Colab Free gpu training, visit mytext version tutorial

To get detailed instructions how to use Yolov3-Tiny, follow my text version tutorialYOLOv3-Tiny support. Short instructions:

• Get YOLOv3-Tiny weights:wget -P model_data https://pjreddie.com/media/files/yolov3-tiny.weights
• Fromyolov3/configs.py changeTRAIN_YOLO_TINY fromFalse toTrue
• Rundetection_demo.py script.

To learn more about Object tracking with Deep SORT, visitFollowing link.Quick test:

• Clone this repository;
• Make sure object detection works for you;
• Run object_tracking.py script

YOLO FPS on COCO 2017 Dataset:

TensorRT FPS on COCO 2017 Dataset:

mAP on COCO 2017 Dataset:

TensorRT mAP on COCO 2017 Dataset:

I will give two examples, both will be for YOLOv4 model,quantize_mode=INT8 and model input size will be 608. Detailed tutorial is on thislink.

• Download weights from links above;
• Inconfigs.py script choose yourYOLO_TYPE;
• Inconfigs.py script setYOLO_INPUT_SIZE = 608;
• Inconfigs.py script setYOLO_FRAMEWORK = "trt";
• From main directory in terminal typepython tools/Convert_to_pb.py;
• From main directory in terminal typepython tools/Convert_to_TRT.py;
• Inconfigs.py script setYOLO_CUSTOM_WEIGHTS = f'checkpoints/{YOLO_TYPE}-trt-{YOLO_TRT_QUANTIZE_MODE}–{YOLO_INPUT_SIZE}';
• Now you can rundetection_demo.py, best to test withdetect_video function.

• Download weights from links above;
• Inconfigs.py script choose yourYOLO_TYPE;
• Inconfigs.py script setYOLO_INPUT_SIZE = 608;
• Train custom YOLO model with instructions above;
• Inconfigs.py script setYOLO_CUSTOM_WEIGHTS = f"{YOLO_TYPE}_custom";
• Inconfigs.py script make sure thatTRAIN_CLASSES is with your custom classes text file;
• From main directory in terminal typepython tools/Convert_to_pb.py;
• From main directory in terminal typepython tools/Convert_to_TRT.py;
• Inconfigs.py script setYOLO_FRAMEWORK = "trt";
• Inconfigs.py script setYOLO_CUSTOM_WEIGHTS = f'checkpoints/{YOLO_TYPE}-trt-{YOLO_TRT_QUANTIZE_MODE}–{YOLO_INPUT_SIZE}';
• Now you can rundetection_custom.py, to test custom trained and converted TensorRT model.

• Detection with original weightsTutorial link
• Mnist detection trainingTutorial link
• Custom detection trainingTutorial link1,link2
• Google Colab trainingTutorial link
• YOLOv3-Tiny supportTutorial link
• Object trackingTutorial link
• Mean Average Precision (mAP)Tutorial link
• Yolo v3 on Raspberry PiTutorial link
• YOLOv4 and YOLOv4-tiny detectionTutorial link
• YOLOv4 and YOLOv4-tiny detection training (Not fully)Tutorial link
• Convert to TensorRT modelTutorial link
• Add multiprocessing after detection (drawing bbox)Tutorial link
• Generate YOLO Object Detection training data from its own resultsTutorial link
• Counter-strike Global Offensive realtime YOLOv4 Object Detection aimbotTutorial link

• Converting to TensorFlow Lite
• YOLO on Android (Leaving it for future, will need to convert everythin to java... not ready for this)
• Generating anchors
• YOLACT: Real-time Instance Segmentation
• Model pruning (Pruning is a technique in deep learning that aids in the development of smaller and more efficient neural networks. It's a model optimization technique that involves eliminating unnecessary values in the weight tensor.)