A Tensorflow Implementation of Yolov3.
- Detecting objects from pretrained coco weights or our trained model.
- Yolov3 training.
- Fine-tuning or Training from scratch.
- Metrics mAP.
- Training process optimization.
-
pretrained darknet weights
The pretrained darknet weights file can be downloaded here. Place this weights file under directory ./data and then run:
python convert_weights.py
-
anchors
Put anchors in the ./data/anchors.txt
10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326
After preparation, we will get:
|--data
|--yolov3.weights
|--checkpoints
|--checkpoint
|--yolov3.ckpt.data-00000-of-00001
|--yolov3.ckpt.index
|--yolov3.ckpt.meta
|--anchors.txt
-
use coco-trained model
python detect.py --image_path utils/COCO_test2014_000000000069.jpg
or
-
use our trained model
python detect.py --image_path utils/2008_002047.jpg --ckpt_path ./checkpoints --names_path ./utils/voc.names --num_classes 20
-
train from pretrained coco model
python train.py
or
-
train from our checkpoint model
python train.py --restore_ckpt_path ./chcekpoints