Easy training on custom dataset. Various backends (MobileNet and SqueezeNet) supported. A YOLO demo to detect raccoon run entirely in brower is accessible at https://git.io/vF7vI (not on Windows).
I've a question regarding the 'pretrained weights'. If I have trained a model for 40 epoches and saved the best to xxx.h5 model，if I want to continue using this model to train and find whether the mAP can raise more, can I just write 'xxx.h5' to the item 'pretrained_weights' in the config.json? Will the parameters of xxx.h5 loaded in my new training model? Thanks a lot.
jzx-gooner
commented
5 years ago
Dear @experiencor and ladys and gentlemen,
I've a question regarding the 'pretrained weights'. If I have trained a model for 40 epoches and saved the best to xxx.h5 model，if I want to continue using this model to train and find whether the mAP can raise more, can I just write 'xxx.h5' to the item 'pretrained_weights' in the config.json? Will the parameters of xxx.h5 loaded in my new training model? Thanks a lot.