context-encoder icon indicating copy to clipboard operation
context-encoder copied to clipboard

Download random region inpainting model

Open cryptedp opened this issue 8 years ago • 8 comments

Hi In your paper you mentioned that you have implemented the 3 strategies, namely centre inpainting, random block inpainting and random region inpainting. However there is currently no way to download a inpaintRandomRegin model in t7 format. Can that be provided or does it need to be trained with train_random.lua?

cryptedp avatar May 10 '17 07:05 cryptedp

Hi @cryptedp ! Yes, Sorry about that. I will have to retrain the train_random.lua and upload the model at some point. I will put on my todo list to run it. However, you can run train own model using train_random.lua on imagenet. Training should be super easy to launch as long as you have Torch and GPU. :)

pathak22 avatar May 10 '17 18:05 pathak22

Hi @pathak22 , how can one train it on imagenet and how long would it take ?

cryptedp avatar May 28 '17 17:05 cryptedp

@pathak22 can you provide already trained model for random region inpainting ?

maryam089 avatar Nov 09 '17 02:11 maryam089

@maryam089 @pathak22 does any one of you have a trained model for random region inpainting?

niloleart avatar Jan 16 '18 18:01 niloleart

Well I spend 6 days and train the model but results were very bad for random region inpaint

maryam089 avatar Jan 16 '18 18:01 maryam089

@maryam089 can you post some results here?

pathak22 avatar Feb 17 '18 08:02 pathak22

@maryam089 Also see #22 (the second picture). Are your results worse than those?

pathak22 avatar Feb 17 '18 08:02 pathak22

I trained your network from strach for random region inpaint at input size 128x128 ... on imageNet 100k ... and results weren’t good then I realized that I should again train it on 64x64 size input.... but as it takes 6 days to train ... so can you provide already trained network of random region inpaint on input size 64x64 ... if possible?

maryam089 avatar Feb 17 '18 08:02 maryam089