Skip to content
Branch: master
Find file History
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
..
Failed to load latest commit information.
README.md
darknet2npz.py
demo.py

README.md

Examples of YOLO [1, 2]

Performance

PASCAL VOC2007 Test

Model Original ChainerCV (weight conversion)
YOLOv2 75.8 % * 75.8 %
YOLOv2 tiny 54.0 % ** 53.5 %
YOLOv3 80.2 % 80.2 %

Scores are mean Average Precision (mAP) with PASCAL VOC2007 metric.

*: Although the original paper [1] reports 76.8 %, the darknet implementation and the provided weights achieved the lower score. Similar issue is reported here.
**: Although the author's website reports 57.1 %, the darknet implementation and the provided weights achieved the lower score.

Demo

Detect objects in an given image. This demo downloads Pascal VOC pretrained model automatically if a pretrained model path is not given.

$ python demo.py [--model yolo_v2|yolo_v2_tiny|yolo_v3] [--gpu <gpu>] [--pretrained-model <model_path>] <image>.jpg

Convert Darknet model

Convert *.weights to *.npz. YOLOv2, YOLOv2 tiny, and YOLOv3 are supported. Note that the number of classes should be specified if it is not 80 (the number of classes in COCO).

$ python darknet2npz.py [--model yolo_v2|yolo_v2_tiny|yolo_v3] [--n-fg-class <#fg_class>]  <source>.weights <target>.npz

Evaluation

The evaluation can be conducted using chainercv/examples/detection/eval_detection.py.

References

  1. Joseph Redmon et al. "YOLO9000: Better, Faster, Stronger" CVPR 2017.
  2. Joseph Redmon et al. "YOLOv3: An Incremental Improvement" arXiv 2018.
You can’t perform that action at this time.