DW2TF: Darknet to TensorFlow
This is a simple converter which converts:
- Darknet weights (
.weights) to TensorFlow weights (.ckpt) - Darknet model (
.cfg) to TensorFlow graph (.pb,.meta)
Requirements
- Ubuntu
- Python 3.6 (known issues with Python 3.7)
Use it
For a full list of options:
python3 main.py -h
Provide optional argument --training to generate training graph (uses batch norm in training mode).
Object Detection Networks
yolov2
python3 main.py \
--cfg 'data/yolov2.cfg' \
--weights 'data/yolov2.weights' \
--output 'data/' \
--prefix 'yolov2/' \
--gpu 0
yolov2-tiny
python3 main.py \
--cfg 'data/yolov2-tiny.cfg' \
--weights 'data/yolov2-tiny.weights' \
--output 'data/' \
--prefix 'yolov2-tiny/' \
--gpu 0
yolov3
python3 main.py \
--cfg 'data/yolov3.cfg' \
--weights 'data/yolov3.weights' \
--output 'data/' \
--prefix 'yolov3/' \
--gpu 0
yolov3-tiny
python3 main.py \
--cfg 'data/yolov3-tiny.cfg' \
--weights 'data/yolov3-tiny.weights' \
--output 'data/' \
--prefix 'yolov3-tiny/' \
--gpu 0
Image Classification Networks
darknet19
python3 main.py \
--cfg 'data/darknet19.cfg' \
--weights 'data/darknet19.weights' \
--output 'data/' \
--prefix 'darknet19/' \
--gpu 0
darknet19_448
python3 main.py \
--cfg 'data/darknet19_448.cfg' \
--weights 'data/darknet19_448.weights' \
--output 'data/' \
--prefix 'darknet19_448/' \
--gpu 0
Todo
- More layer types