Closed sukkritsharmaofficial closed 3 years ago
I find this PR very misleading ,there is nothing called as "Deeper models" and "Lightweight models".The key difference is with respect to the number of parameters a model is using.What i inferred from this you are just trying to copy and paste the model weights that are provided by keras .If that is so any kind of DCNN models can be added without taking in account the use case.
Firstly its not a PR, its just an issue that I've raised, but since you have joined Github half an hour back I can understand your confusion. Talking about the issue raised, since obviously you haven't used the Library as such and have not gone through the whole code, you wouldn't understand that this library uses models from keras and while training your model you are given two choices, one is to use pretrained models with weights from imagenet or custom weights or using a custom architecture defined in a json file. But since they had added very less number of models as of now, I have raised this issue. Its anyways better than editing ReadMe and calling it as a PR.
this is not the place to quarrel. on the other hand @sukkritsharmaofficial that sounds good!
By the way, if you've got any questions about implementation etc. you can join the slack channel and PM me.
@Palashio yeah sure, i'll hit you up there!
I can help you out with adding support to more pre-trained models which are usually considered while training an image classifier
Light Weight models like:
Substantially Deeper Models like:
Adding support for these will give the users a wide variety while selecting the architectures that they need to train on depending upon their specific use case 💯