A yolov3 simple implementation in MXNet, based on version 1.2.0 and cuda 9.0(optional), python3. Works on Windows and Ubuntu 16.04.
**new: hybridized, speed up.**
**new: train demo.**
**Detect Part Completed.**
=======
Joseph Redmon, Ali Farhadi <br >
Abstract <br > 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]
$ git clone https://github.com/Fermes/yolov3-mxnet.git
$ cd yolov3-mxnet/
$ sudo pip3 install opencv-python mxnet-cu90==1.2.0
put your images in ./images, and
$ python detect.py [--gpu GPU ID]
you will get the results in ./results or you can detect a video file, this need your opencv is compiled with ffmpeg.
$ python detect.py --video VIDEO_FILE
you will get result.avi in ./results
The IMAGE_FOLDER should contains two directorys, "train" and "train_label", (or four, "train", "train_label", "val", "val_label") label should be xml file like VOC's format. PS: You can use voc_label.py in https://pjreddie.com/darknet/yolo/ to get train.txt and val.txt, set path to --train and --val instead of --images.
train.py [-h] [--epochs EPOCHS] [--images IMAGES FOLDER]
[--batch_size BATCH_SIZE]
[--params WEIGHTS_PATH] [--classes CLASS_PATH]
[--confidence CONF_THRES] [--nms_thresh NMS_THRES]
[--gpu GPU ID] [--prefix PARAMS FILE NAME PREFIX]
@article{yolov3,
title={YOLOv3: An Incremental Improvement},
author={Redmon, Joseph and Farhadi, Ali},
journal = {arXiv},
year={2018}
}