FreeYOLO

Anchor-free YOLO detector.

Requirements

• We recommend you to use Anaconda to create a conda environment:

conda create -n yolo python=3.6

• Then, activate the environment:

conda activate yolo

• Requirements:

pip install -r requirements.txt 

My environment:

• PyTorch = 1.9.1
• Torchvision = 0.10.1

At least, please make sure your torch is version 1.x.

Tricks

• Mosaic Augmentation
• Mixup Augmentation
• Multi scale training
• Cosine Annealing Schedule

Training Configuration

Configuration
Batch Size (bs)	16
Init Lr	0.01/64 × bs
Lr Scheduler	Cos
Optimizer	SGD
ImageNet Predtrained	True
Multi Scale Train	True
Mosaic	True
Mixup	True

Experiments

COCO

• Download COCO.

cd <FreeYOLO_HOME>
cd dataset/scripts/
sh COCO2017.sh

• Check COCO

cd <FreeYOLO_HOME>
python dataset/coco.py

• Train on COCO

For example:

python train.py --cuda -d coco -v yolo_free_nano -bs 16 -accu 4 --max_epoch 300 --wp_epoch 1 --eval_epoch 10 --fp16 --ema --root path/to/COCO

Main results on COCO-val:

Model	Scale	FPS 2080ti	FLOPs	Params	AP	AP50	Weight
FreeYOLO-Nano	640		4.6 G	2.0 M	30.5	50.3	github
FreeYOLO-Tiny	640		13.9 G	6.2 M	34.4	53.9	github
FreeYOLO-Large	640		144.8 G	44.1 M	48.3	68.5	github
FreeYOLO-Huge	640		257.8 G	78.9 M

WiderFace

• Download WiderFace.

• Prepare WiderFace

WiderFace
|_ WIDER_train
|  |_ images
|     |_ 0--Parade
|     |_ ...
|_ WIDER_tval
|  |_ images
|     |_ 0--Parade
|     |_ ...
|_ wider_face_split
|_ eval_tools

• Convert WiderFace to COCO format.

cd <FreeYOLO_HOME>
python tools/convert_widerface_to_coco.py --root path/to/WiderFace

• Check WiderFace

cd <FreeYOLO_HOME>
python dataset/widerface.py

• Train on WiderFace For example:

python train.py --cuda -d widerface -v yolo_free_nano -bs 16 -accu 4 --max_epoch 100 --wp_epoch 1 --eval_epoch 10 --fp16 --ema --root path/to/WiderFace --pretrained path/to/coco/yolo_free_nano_ckpt --mosaic 0.5, --mixup 0.0

Main results on WiderFace-val:

Model	Scale	AP	AP50	Weight
FreeYOLO-Nano	640
FreeYOLO-Tiny	640
FreeYOLO-Large	640
FreeYOLO-Huge	640

CrowdHuman

• Download CrowdHuman.

CrowdHuman
|_ CrowdHuman_train01.zip
|_ CrowdHuman_train02.zip
|_ CrowdHuman_train03.zip
|_ CrowdHuman_val.zip
|_ annotation_train.odgt
|_ annotation_val.odgt

• Prepare CrowdHuman

CrowdHuman
|_ CrowdHuman_train
|  |_ Images
|     |_ 273271,1a0d6000b9e1f5b7.jpg
|     |_ ...
|_ CrowdHuman_val
|  |_ Images
|     |_ 273271,1b9330008da38cd6.jpg
|     |_ ...
|_ annotation_train.odgt
|_ annotation_val.odgt

• Convert CrowdHuman to COCO format.

cd <FreeYOLO_HOME>
python tools/convert_crowdhuman_to_coco.py --root path/to/CrowdHuman

• Check CrowdHuman

cd <FreeYOLO_HOME>
python dataset/crowdhuman.py

• Train on CrowdHuman For example:

python train.py --cuda -d crowdhuman -v yolo_free_nano -bs 16 -accu 4 --max_epoch 100 --wp_epoch 1 --eval_epoch 10 --fp16 --ema --root path/to/CrowdHuman --pretrained path/to/coco/yolo_free_nano_ckpt

Main results on CrowdHuman-val:

Model	Scale	AP	AP50	Weight
FreeYOLO-Nano	640	31.3	67.2	github
FreeYOLO-Tiny	640	34.7	70.4	github
FreeYOLO-Large	640	43.1	76.5	github
FreeYOLO-Huge	640

MOT17

• Download MOT17, then you will get a ```MOT17.zip` file.

• Prepare MOT17

MOT17
|_ train
|  |_ MOT17-02-DPM
|     |_ det
|     |_ gt
|     |_ img1
|     |_ ...
|  ...
|_ test
|  |_ MOT17-01-DPM
|     |_ det
|     |_ img1
|     |_ ...
|  ...

• Convert MOT17 to COCO format.

cd <FreeYOLO_HOME>
python tools/convert_mot17_to_coco.py --root path/to/MOT17

• Check MOT17

cd <FreeYOLO_HOME>
python dataset/mot17.py

• Train on MOT17 half

For example:

python train.py --cuda -d mot17_half -v yolo_free_nano -bs 16 -accu 4 --max_epoch 100 --wp_epoch 1 --eval_epoch 10 --fp16 --ema --root path/to/MOT17 --pretrained path/to/coco/yolo_free_nano_ckpt

Main results on MOT17 val-half (trained on MOT17 train-half):

Model	Scale	AP	AP50	Weight
FreeYOLO-Nano	640
FreeYOLO-Tiny	640
FreeYOLO-Large	640
FreeYOLO-Huge	640

• Train on MOT17

For example:

python train.py --cuda -d mot17 -v yolo_free_nano -bs 16 -accu 4 --max_epoch 100 --wp_epoch 1 --fp16 --ema --root path/to/MOT17 --pretrained path/to/coco/yolo_free_nano_ckpt

Pretrained weights on MOT17 train split (fully train, not train-half):

FreeYOLO-Nano

FreeYOLO-Tiny

FreeYOLO-Large

FreeYOLO-Huge

MOT20

• Download MOT20, then you will get a ```MOT20.zip` file.

• Prepare MOT20 Similar to MOT17

• Convert MOT20 to COCO format.

cd <FreeYOLO_HOME>
python tools/convert_mot20_to_coco.py --root path/to/MOT20

• Check MOT20

cd <FreeYOLO_HOME>
python dataset/mot20.py

• Train on MOT20 half

For example:

python train.py --cuda -d mot20_half -v yolo_free_nano -bs 16 --max_epoch 100 --wp_epoch 1 --eval_epoch 10 --fp16 --ema --root path/to/MOT20 --pretrained path/to/coco/yolo_free_nano_ckpt

Main results on MOT20 val-half (trained on MOT20 train-half):

Model	Scale	AP	AP50	Weight
FreeYOLO-Nano	640
FreeYOLO-Tiny	640
FreeYOLO-Large	640
FreeYOLO-Huge	640

• Train on MOT20

For example:

python train.py --cuda -d mot20 -v yolo_free_nano -bs 16 --max_epoch 100 --wp_epoch 1 -- eval_epoch 10 --fp16 --ema --root path/to/MOT20 --pretrained path/to/coco/yolo_free_nano_ckpt

Pretrained weights on MOT20 train split (fully train, not train-half):

FreeYOLO-Nano

FreeYOLO-Tiny

FreeYOLO-Large

FreeYOLO-Huge

Train

Single GPU

sh train.sh

You can change the configurations of train.sh, according to your own situation.

Multi GPUs

sh train_ddp.sh

You can change the configurations of train_ddp.sh, according to your own situation.

In the event of a training interruption, you can pass --resume the latest training weight path (None by default) to resume training. For example:

python train.py \
        --cuda \
        -d coco \
        -v yolo_free_large \
        --ema \
        --fp16 \
        --eval_epoch 10 \
        --resume weights/coco/yolo_free_large/yolo_free_large_epoch_151_39.24.pth

Then, training will continue from 151 epoch.

Test

python test.py -d coco \
               --cuda \
               -v yolo_free_large \
               --img_size 640 \
               --weight path/to/weight \
               --root path/to/dataset/ \
               --show

Evaluation

python eval.py -d coco-val \
               --cuda \
               -v yolo_free_large \
               --img_size 640 \
               --weight path/to/weight \
               --root path/to/dataset/ \
               --show

Demo

I have provide some images in data/demo/images/, so you can run following command to run a demo:

python demo.py --mode image \
               --path_to_img data/demo/images/ \
               -v yolo_free_large \
               --img_size 640 \
               --cuda \
               --weight path/to/weight

If you want run a demo of streaming video detection, you need to set --mode to video, and give the path to video --path_to_vid。

python demo.py --mode video \
               --path_to_img data/demo/videos/your_video \
               -v yolo_free_large \
               --img_size 640 \
               --cuda \
               --weight path/to/weight

If you want run video detection with your camera, you need to set --mode to camera。

python demo.py --mode camera \
               -v yolo_free_large \
               --img_size 640 \
               --cuda \
               --weight path/to/weight

Deployment

1. ONNX export and an ONNXRuntime
2. OpenVINO in C++ and Python