search

SeungjunNah / DeepDeblur-PyTorch

Deep Multi-scale CNN for Dynamic Scene Deblurring
MIT License
262 stars 42 forks source link

[Bug] Expected Demo Input Format #19

Closed mohammed-amr closed 4 years ago

mohammed-amr commented 4 years ago

I have some jpg images I'd like to run the demo on, but the output seems to be erroneous in some way.

Here's an example input image: image

Here's an example output image: image

I followed the instructions for the first demo and have not changed anything in the code. Using Pytorch 1.6, torchvision 0.7.0, and Python 3.6. Using the GOPRO_L1 pretrained experiment weights. Is there an assumption on type of image and range?

mohammed-amr commented 4 years ago

Changing the blur camera response type to blur from the default of blur_gamma doesn't produce a better result.

image

mohammed-amr commented 4 years ago

Fixed my own issue with a solution mentioned elsewhere. Seems like the codebase fails silently when it can't find the right model weights. Set a full explicit path instead of a relative path and it should be ok.