search

david8862 / keras-YOLOv3-model-set

end-to-end YOLOv4/v3/v2 object detection pipeline, implemented on tf.keras with different technologies
MIT License
639 stars 220 forks source link

Do you use the darknet *.cfg files for anything other than conversion of .weights to .h5? #69

Open bsugerman opened 4 years ago

bsugerman commented 4 years ago

Are the darknet .cfg files that define models used for anything other than conversion of darknet .weights to keras .h5? It seems overly complicated to need to use the cfg files and the models defined natively within keras. There are some keras version of yolo that seem to use the .weights files directly, rather than needing to convert them, such as

https://github.com/experiencor/keras-yolo3/blob/master/yolo3_one_file_to_detect_them_all.py

Figured it is worth bringing that to your attention.

david8862 commented 4 years ago

Many thanks for suggestion. All the .cfg files in this repo are from official darknet code base and currently their purpose are only converting darknet pretrained .weights to keras .h5 model file, in order to quickly model verification. Normal model construction here will still use keras native code, so that we can leverage rich types of backbone implementation in keras/tf.keras and their official pretrained weights, to easily build different YOLO model structure.