wolframrhodium / super-resolution-zoo

Collection of pre-trained super-resolution models in MXNet.

Super-Resolution-Zoo

Collection of pre-trained super-resolution models in MXNet.

There are also some models for other low-level vision tasks, e.g. denoising and deblocking.

Notice: I DO NOT guarantee that models are converted perfectly. I do have checked some of them (e.g. EDSR, RCAN), but most of the other models have only been tested roughly. Please feel free to report any issues that looks unnatural.

Demo@sa-mustafa

Explanation of configurations in info.md

super-resolution-zoo's Issues

Add the framework of the model in the individual descriptions

Some people don't know which framework was used just by looking at the file extensions and some may even be not too easy to find due generic naming (.bin, .params, .json) adding would help newer people.

Add ONNX version of the models

The new standard ONNX allows for interchangeable models, adding this format to the zoo would allow the models to be converted and used with ease.

https://onnx.ai/

Question about RCAN

Hi,

I try to run the Super Resolution with RCAN method with RCAN_BDX3-0000 params and json.
My input image has a 224x224 dimension. I load the model as following:

symbol, arg_params, aux_params = mx.model.load_checkpoint('RCAN_BDX3', 0)
model = mx.mod.Module(symbol=symbol, label_names=None)
model.bind(data_shapes=[('data', (1, 3, 224, 224))], for_training=False)
model.init_params(arg_params=arg_params, aux_params=aux_params, allow_missing=False, allow_extra=False)

Then I load the image:
img = mx.image.imread(fname)
img_batch = DataBatch(mx.nd.array(img), 0)

And forward the model:
model.forward(img_batch)

The forward method generates the following error:
MXNetError: Error in operator pre_mean_shift: [14:55:42] c:\jenkins\workspace\mxnet-tag\mxnet\src\operator\tensor\./elemwise_binary_broadcast_op.h:68: Check failed: l == 1 || r == 1 operands could not be broadcast together with shapes [224,224,3] [1,3,1,1]

Does there is any modification to do inside JSON file ("attr": {"__shape__": "(1, 3, 1, 1)"} for instance) to run the model?

Thank you

waifu2x bug

using kice's VapourSynth plugin, the edge is broken at many places. http://puu.sh/Di5wy/7a6f011f80.jpg http://puu.sh/Di5zC/7fbcf2c16c.jpg
the old models from around October do not have this issue. But the old ones requires adding paddings for scaling. The new waifu2x models (lltcggie) do not require paddings for scaling but they are all having broken edges.

How to use these models?

Would you please release one demo.py about how to use models?

MoePhoto official dehazing result have some white line with mx.Predict.

http://www.imagebam.com/image/4d46c41015403564

Also How to Converting pre-trained pytorch .pkl model to MXNet?(Try MMdnn few hours but no luck)

Issue with Strides in RED30

Hi ,

In this repository ,

https://github.com/WolframRhodium/Super-Resolution-Zoo/blob/master/RED30/super_resolution/RED30_4x-symbol.json

has stride in all convolution / deconvolution layers as *** "stride": "(1,1)" ***
Can you please let me know if it is an error and suggest the right strides for the layers

Awaiting a speedy response,

Model performance comparision

A table showing the results for each algorithm for the same photos would be a great addition to this repo!