Open VNVetrov opened 12 months ago
Well, I did a test and I failed I guess
This is the result I got on 3k iteration and its not getting better I put the first image as an input image As a mask I put a mask of top right watermark
Right image is a result and I think I did something wrong
About inference time: on my specs (see above) it took 1 minute on GPU with such params: show_step=3000, max_dim=128, reg_noise=0.03, input_depth=64, lr=0.01, training_steps=10000
@braindotai If I'm doing something wrong pls lemme know
I have to look into it, I think model is not performing well for all black input... I gotta test a bit and look into its potential solution. Thanks for trying my project out and highlighting one of the current potential edge case.
I have to look into it, I think model is not performing well for all black input... I gotta test a bit and look into its potential solution. Thanks for trying my project out and highlighting one of the current potential edge case.
Oh, The model performs poorly with a white mask on a black background, it cannot remove the white mask from the black background
Really nice project! Thank you for making it public!
The question is: what's the time of inference on image? Actually my goal is to remove watermark in the bottom rigth corner of the video. I've already prepared the docker container and simple code to divide video into frames. How long will it take to run on 1k frames?
I can't actually check for now because video is not currently available. (I'll post my results when I finish)
Specs: