Skip to content
Minimal PyTorch implementation of YOLOv3
Branch: master
Clone or download
Latest commit ac7bb33 Apr 20, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
assets Removed output folder and replaced it with assets folder, which conta… May 22, 2018
config Now tiny yolo-v3 can be used as detecting network, which is lightweig… Jul 10, 2018
data
utils Tensorboard Apr 19, 2019
weights Now tiny yolo-v3 can be used as detecting network, which is lightweig… Jul 10, 2018
.gitignore
README.md Tensorboard Apr 19, 2019
detect.py Added argument use_cuda Jun 13, 2018
models.py Tensorboard Apr 19, 2019
requirements.txt Tensorboard Apr 19, 2019
test.py Added precision to train log Oct 1, 2018
train.py Tensorboard Apr 19, 2019

README.md

PyTorch-YOLOv3

Minimal implementation of YOLOv3 in PyTorch.

Table of Contents

Paper

YOLOv3: An Incremental Improvement

Joseph Redmon, Ali Farhadi

Abstract
We present some updates to YOLO! We made a bunch of little design changes to make it better. We also trained this new network that’s pretty swell. It’s a little bigger than last time but more accurate. It’s still fast though, don’t worry. At 320 × 320 YOLOv3 runs in 22 ms at 28.2 mAP, as accurate as SSD but three times faster. When we look at the old .5 IOU mAP detection metric YOLOv3 is quite good. It achieves 57.9 AP50 in 51 ms on a Titan X, compared to 57.5 AP50 in 198 ms by RetinaNet, similar performance but 3.8× faster. As always, all the code is online at https://pjreddie.com/yolo/.

[Paper] [Original Implementation]

Installation

$ git clone https://github.com/eriklindernoren/PyTorch-YOLOv3
$ cd PyTorch-YOLOv3/
$ sudo pip3 install -r requirements.txt
Download pretrained weights
$ cd weights/
$ bash download_weights.sh
Download COCO
$ cd data/
$ bash get_coco_dataset.sh

Inference

Uses pretrained weights to make predictions on images. Below table displays the inference times when using as inputs images scaled to 256x256. The ResNet backbone measurements are taken from the YOLOv3 paper. The Darknet-53 measurement marked shows the inference time of this implementation on my 1080ti card.

Backbone GPU FPS
ResNet-101 Titan X 53
ResNet-152 Titan X 37
Darknet-53 (paper) Titan X 76
Darknet-53 (this impl.) 1080ti 74
$ python3 detect.py --image_folder /data/samples

Test

Evaluates the model on COCO test.

$ python3 test.py --weights_path weights/yolov3.weights
Model mAP (min. 50 IoU)
YOLOv3 (paper) 57.9
YOLOv3 (this impl.) 58.2

Train

Data augmentation as well as additional training tricks remains to be implemented. PRs are welcomed!

    train.py [-h] [--epochs EPOCHS] [--image_folder IMAGE_FOLDER]
                [--batch_size BATCH_SIZE]
                [--model_config_path MODEL_CONFIG_PATH]
                [--data_config_path DATA_CONFIG_PATH]
                [--weights_path WEIGHTS_PATH] [--class_path CLASS_PATH]
                [--n_cpu N_CPU] [--img_size IMG_SIZE]
                [--checkpoint_model CHECKPOINT_MODEL]
                [--checkpoint_interval CHECKPOINT_INTERVAL]
                [--checkpoint_dir CHECKPOINT_DIR]

TensorBoard

Track training progress in Tensorboard:

$ tensorboard --logdir='logs' --port=6006

Credit

@article{yolov3,
  title={YOLOv3: An Incremental Improvement},
  author={Redmon, Joseph and Farhadi, Ali},
  journal = {arXiv},
  year={2018}
}
You can’t perform that action at this time.