YOLOR
implementation of paper - You Only Learn One Representation: Unified Network for Multiple Tasks
To get the results on the table, please use this branch.
| Model | Test Size | APtest | AP50test | AP75test | batch1 throughput | batch32 inference |
|---|---|---|---|---|---|---|
| YOLOR-P6 | 1280 | 54.1% | 71.8% | 59.3% | 76 fps | 8.3 ms |
| YOLOR-W6 | 1280 | 55.5% | 73.2% | 60.6% | 66 fps | 10.7 ms |
| YOLOR-E6 | 1280 | 56.4% | 74.1% | 61.6% | 45 fps | 17.1 ms |
| YOLOR-D6 | 1280 | 57.3% | 75.0% | 62.7% | 34 fps | 21.8 ms |
| YOLOR-D6* | 1280 | 57.8% | 75.5% | 63.3% | 34 fps | 21.8 ms |
| YOLOv4-P5 | 896 | 51.8% | 70.3% | 56.6% | 41 fps (old) | - |
| YOLOv4-P6 | 1280 | 54.5% | 72.6% | 59.8% | 30 fps (old) | - |
| YOLOv4-P7 | 1536 | 55.5% | 73.4% | 60.8% | 16 fps (old) | - |
- Fix the speed bottleneck on our NFS, many thanks to NCHC, TWCC, and NARLabs support teams.
To reproduce the inference speed, please see darknet.
| Model | Test Size | APval | AP50val | AP75val | APSval | APMval | APLval | batch1 throughput |
|---|---|---|---|---|---|---|---|---|
| YOLOv4-CSP | 640 | 49.1% | 67.7% | 53.8% | 32.1% | 54.4% | 63.2% | 76 fps |
| YOLOR-CSP | 640 | 49.2% | 67.6% | 53.7% | 32.9% | 54.4% | 63.0% | weights |
| YOLOR-CSP* | 640 | 50.0% | 68.7% | 54.3% | 34.2% | 55.1% | 64.3% | weights |
| YOLOv4-CSP-X | 640 | 50.9% | 69.3% | 55.4% | 35.3% | 55.8% | 64.8% | 53 fps |
| YOLOR-CSP-X | 640 | 51.1% | 69.6% | 55.7% | 35.7% | 56.0% | 65.2% | weights |
| YOLOR-CSP-X* | 640 | 51.5% | 69.9% | 56.1% | 35.8% | 56.8% | 66.1% | weights |
Developing...
| Model | Test Size | APtest | AP50test | AP75test | APStest | APMtest | APLtest |
|---|---|---|---|---|---|---|---|
| YOLOR-CSP | 640 | 51.1% | 69.6% | 55.7% | 31.7% | 55.3% | 64.7% |
| YOLOR-CSP-X | 640 | 53.0% | 71.4% | 57.9% | 33.7% | 57.1% | 66.8% |
Train from scratch for 300 epochs...
| Model | Info | Test Size | AP |
|---|---|---|---|
| YOLOR-CSP | evolution | 640 | 48.0% |
| YOLOR-CSP | strategy | 640 | 50.0% |
| YOLOR-CSP | strategy + simOTA | 640 | 51.1% |
| YOLOR-CSP-X | strategy | 640 | 51.5% |
| YOLOR-CSP-X | strategy + simOTA | 640 | 53.0% |
Installation
Docker environment (recommended)
Expand
# create the docker container, you can change the share memory size if you have more.
nvidia-docker run --name yolor -it -v your_coco_path/:/coco/ -v your_code_path/:/yolor --shm-size=64g nvcr.io/nvidia/pytorch:20.11-py3
# apt install required packages
apt update
apt install -y zip htop screen libgl1-mesa-glx
# pip install required packages
pip install seaborn thop
# install mish-cuda if you want to use mish activation
# https://github.com/thomasbrandon/mish-cuda
# https://github.com/JunnYu/mish-cuda
cd /
git clone https://github.com/JunnYu/mish-cuda
cd mish-cuda
python setup.py build install
# install pytorch_wavelets if you want to use dwt down-sampling module
# https://github.com/fbcotter/pytorch_wavelets
cd /
git clone https://github.com/fbcotter/pytorch_wavelets
cd pytorch_wavelets
pip install .
# go to code folder
cd /yolor
Colab environment
Expand
git clone https://github.com/WongKinYiu/yolor
cd yolor
# pip install required packages
pip install -qr requirements.txt
# install mish-cuda if you want to use mish activation
# https://github.com/thomasbrandon/mish-cuda
# https://github.com/JunnYu/mish-cuda
git clone https://github.com/JunnYu/mish-cuda
cd mish-cuda
python setup.py build install
cd ..
# install pytorch_wavelets if you want to use dwt down-sampling module
# https://github.com/fbcotter/pytorch_wavelets
git clone https://github.com/fbcotter/pytorch_wavelets
cd pytorch_wavelets
pip install .
cd ..
Prepare COCO dataset
Expand
cd /yolor
bash scripts/get_coco.sh
Prepare pretrained weight
Expand
cd /yolor
bash scripts/get_pretrain.sh
Testing
python test.py --data data/coco.yaml --img 1280 --batch 32 --conf 0.001 --iou 0.65 --device 0 --cfg cfg/yolor_p6.cfg --weights yolor_p6.pt --name yolor_p6_val
You will get the results:
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.52510
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.70718
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.57520
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.37058
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.56878
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.66102
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.39181
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.65229
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.71441
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.57755
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.75337
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.84013
Training
Single GPU training:
python train.py --batch-size 8 --img 1280 1280 --data coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0 --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
Multiple GPU training:
python -m torch.distributed.launch --nproc_per_node 2 --master_port 9527 train.py --batch-size 16 --img 1280 1280 --data coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0,1 --sync-bn --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
Training schedule in the paper:
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights '' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 tune.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights 'runs/train/yolor_p6/weights/last_298.pt' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6-tune --hyp hyp.finetune.1280.yaml --epochs 450
python -m torch.distributed.launch --nproc_per_node 8 --master_port 9527 train.py --batch-size 64 --img 1280 1280 --data data/coco.yaml --cfg cfg/yolor_p6.cfg --weights 'runs/train/yolor_p6-tune/weights/epoch_424.pt' --device 0,1,2,3,4,5,6,7 --sync-bn --name yolor_p6-fine --hyp hyp.finetune.1280.yaml --epochs 450
Inference
python detect.py --source inference/images/horses.jpg --cfg cfg/yolor_p6.cfg --weights yolor_p6.pt --conf 0.25 --img-size 1280 --device 0
You will get the results:
Citation
@article{wang2021you,
title={You Only Learn One Representation: Unified Network for Multiple Tasks},
author={Wang, Chien-Yao and Yeh, I-Hau and Liao, Hong-Yuan Mark},
journal={arXiv preprint arXiv:2105.04206},
year={2021}
}