ResNet_RS pre-trained

[api5454]

Hello, and thank you for this excellent work,
But I was wondering about the possibility of obtaining weights
resnet50_weights_tf_dim_ordering_tf_kernels.h5

If there is no way to get it, can you give some advice about training it on data ImageNet,
For your information, I have downloaded resnet50 standard weights from TF-V1 and TF-v2 and the number of layers mismatch appears,
Or this error<..
ValueError: You are trying to load a weight file containing 107 layers into a model with 139 layers.

and,
Can I take advantage of the following checkpoints?? ,

ImageNet Checkpoints from TF.tpu.official

We release configs and checkpoints of the ResNet-RS model family trained on ImageNet in Tensorflow 1.

Model	Input Size	V100 Lat (s)	TPU Lat (ms)	Top-1 Accuracy	Download
ResNet-RS-50	160x160	0.31	70	78.8	config | ckpt
ResNet-RS-101	160x160	0.48	120	80.3	config | ckpt
ResNet-RS-101	192x192	0.70	170	81.2	config | ckpt
ResNet-RS-152	192x192	0.99	240	82.0	config | ckpt
ResNet-RS-152	224x224	1.48	320	82.2	config | ckpt
ResNet-RS-152	256x256	1.76	410	83.0	config | ckpt
ResNet-RS-200	256x256	2.86	570	83.4	config | ckpt
ResNet-RS-270	256x256	3.76	780	83.8	config | ckpt
ResNet-RS-350	256x256	4.72	1100	84.0	config | ckpt
ResNet-RS-350	320x320	8.48	1630	84.2	config | ckpt
ResNet-RS-420	320x320	10.16	2090	84.4	config | ckpt

Thank you and good luck to you anyway

[AmberzzZZ]

as you can see there's a little modification on the network architecture from resnet to resnet_rs(the stem & the se_block).
so it's not possible to load the h5 file directly to this model.
but you can try to load the weights layer by layer(as the work i did in covert* scripts).
currently there's some tf version problem so i'm not able to parse the official ckpts correctly but i'm still working on.
after the work is done the convert script would be pushed :)

[api5454]

Thanks for your advice
I got you
Until this time we will close this..