pytorch-yolov3

About

simple Pytorch implementation of YOLOv3 / YOLOv3-tiny.

Features

• [x]: Support darknet format weights
• [x]: Sample code for infering image and video
• [x]: Sample code for training

Setup

1. Install dependent libraries.

   pip install -r requirements.txt

   If you have already installed older version of Pytorch, update it.

   pip install -U torch torchvision torchaudio

2. Download the official weights.

   ./weights/download_weights.sh

Usage (YOLOv3)

Detect from a single image

python detect_image.py \
    --input data/dog.png \
    --output output \
    --weights weights/yolov3.weights \
    --config config/yolov3_coco.yaml

Detect from images in the dicretory.

python detect_image.py \
    --input data \
    --output output \
    --weights weights/yolov3.weights \
    --config config/yolov3_coco.yaml

Detect from a single video

python detect_video.py \
    --input data/sample.avi \
    --output output \
    --weights weights/yolov3.weights \
    --config config/yolov3_coco.yaml

Train custom dataset

YOLOv3 - 自作データセットで学習する方法について - pystyle

python train_custom.py \
    --dataset_dir custom_dataset \
    --weights weights/darknet53.conv.74 \
    --config config/yolov3_custom.yaml

Evaluate custom dataset

python evaluate_custom.py \
    --dataset_dir custom_dataset \
    --weights train_output/yolov3_final.pth \
    --config config/yolov3_custom.yaml

Usage (YOLOv3-tiny)

Detect from a single image

python detect_image.py \
    --input data/dog.png \
    --output output \
    --weights weights/yolov3-tiny.weights \
    --config config/yolov3tiny_coco.yaml

Detect from images in the dicretory.

python detect_image.py \
    --input data \
    --output output \
    --weights weights/yolov3-tiny.weights \
    --config config/yolov3tiny_coco.yaml

Detect from a single video

python detect_video.py \
    --input data/sample.avi \
    --output output \
    --weights weights/yolov3-tiny.weights \
    --config config/yolov3tiny_coco.yaml

Train custom dataset

Download YOLOv3-tiny pretrained weights.

YOLOv3 - 自作データセットで学習する方法について - pystyle

python train_custom.py \
    --dataset_dir custom_dataset \
    --weights weights/yolov3-tiny.conv.15 \
    --config config/yolov3tiny_custom.yaml

MSCOC benchmark

Setup

1. Download MSCOC 2017 dataset and unzip them.

   wget http://images.cocodataset.org/annotations/annotations_trainval2017.zip
   wget http://images.cocodataset.org/zips/train2017.zip
   wget http://images.cocodataset.org/zips/val2017.zip
   unzip train2017.zip
   unzip val2017.zip
   unzip annotations_trainval2017.zip

   After unzipping, the directory structure should look like

   <dataset_dir>
   |-- annotations
   |-- train2017
   `-- val2017
   

2. Install pycocotools (Cython is required to build pycocotools)

   pip install Cython
   pip install pycocotools

Train MSCOCO dataset

python train_coco.py \
    --dataset_dir /data/COCO \
    --anno_path /data/COCO/annotations/instances_train2017.json \
    --weights weights/darknet53.conv.74 \
    --config config/yolov3_coco.yaml

Evaluate on COCO dataset

python evaluate_coco.py \
    --dataset_dir /data/COCO \
    --anno_path config/instances_val5k.json \
    --weights weights/yolov3.weights \
    --config config/yolov3_coco.yaml

 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.311
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.558
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.313
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.141
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.339
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.457
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.275
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.416
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.437
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.238
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.477
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.603
0.31082299897937926 0.5579624636166972