YOLOv4-Counter-in-TF

YOLOv4 implementation for a counter program, using tensorflow. In future trying to make it faster and lighter to deploy to low hardware systems

Video Analysis for this network, in comparison to human volunteers: https://drive.google.com/drive/u/0/folders/1Es_wIs_X8q_UhNZoQSm_tlonMcuPr2vJ

COCO data set, 80 Classes yolov4 weights file (to be add to checkpoints folder): https://drive.google.com/drive/u/0/folders/1ZDs2xCzyOjOckmNZNUGLE20PumyrOkNr

Disclaimer

This github is still in working. Not so much have been done yet, The purpose is to create a vehicle and people counter and make it appliable for deploy.

Prerequisites

pip install -r requirements.txt
pip install -r requirements-gpu.txt

But instead of doing it on your own computer, every step on this repository will be placed in a google-colab notebook.

Demo

# yolov4
python detect.py --weights ./data/yolov4.weights --framework tf --size 608 --image ./data/madureira.jpg --output result1.png

# yolov4 tflite
python detect.py --weights ./data/yolov4-int8.tflite --framework tflite --size 416 --image ./data/kite.jpg

# yolov4 video - with file output
python detectvideo_output.py --weights ./data/yolov4.weights --framework tf --size 608 --video ./path_to_video_file

# yolov4 counter ROI defined on executing
python detectvideo_counter.py --weights ./data/yolov4.weights --framework tf --size 608 --video ./path_to_video_file

# yolov4 fast counter using tiny weights
python detectvideo_counter.py --weights ./checkpoints/yolov4-tiny-416  --video ./data/road3.mp4 --tiny

Outputs

Obtain a dataset for a custom or class focused train

Some times, if you want to increase your detection speed or add some class to your detection, you should retrain your network. I prepared a colab-notebook with dataset download to acquire OpenImages with proper labels for darknet. Colab Notebook for Dataset Download

Traning your own model

The training performance is not fully reproduced yet, so I recommended to use Alex's Darknet to train your own data, then convert the .weights to tensorflow or tflite.

Convert to tflite

# yolov4
python convert_tflite.py --weights ./data/yolov4.weights --output ./data/yolov4.tflite

# yolov4 quantize float16
python convert_tflite.py --weights ./data/yolov4.weights --output ./data/yolov4-fp16.tflite --quantize_mode float16

# yolov4 quantize int8
python convert_tflite.py --weights ./data/yolov4.weights --output ./data/yolov4-fp16.tflite --quantize_mode full_int8 --dataset ./coco_dataset/coco/val207.txt

Convert to TensorRT

# yolov3
python save_model.py --weights ./data/yolov3.weights --output ./checkpoints/yolov3.tf --input_size 416 --model yolov3
python convert_trt.py --weights ./checkpoints/yolov3.tf --quantize_mode float16 --output ./checkpoints/yolov3-trt-fp16-416

# yolov3-tiny
python save_model.py --weights ./data/yolov3-tiny.weights --output ./checkpoints/yolov3-tiny.tf --input_size 416 --tiny
python convert_trt.py --weights ./checkpoints/yolov3-tiny.tf --quantize_mode float16 --output ./checkpoints/yolov3-tiny-trt-fp16-416

# yolov4
python save_model.py --weights ./data/yolov4.weights --output ./checkpoints/yolov4.tf --input_size 416 --model yolov4
python convert_trt.py --weights ./checkpoints/yolov4.tf --quantize_mode float16 --output ./checkpoints/yolov4-trt-fp16-416

References

This project is totally based on the following implementations and Ideas They are amazing!

• darknet
• tfyolov4lite
• TheAIGuy
• Object counter in Yolo