search

posterllava / PosterLLaVA

Other
55 stars 1 forks source link

Would you like to provide the script for generate the saliency_map, inpainted_1x, inpainted_1d5x #3

Closed trouble-maker007 closed 3 months ago

trouble-maker007 commented 4 months ago

use the similar method to get the saliency map and inpainted result: https://github.com/CyberAgentAILab/RALF/tree/c51db6032acbd0bd0ce72433becce08317e7874d?tab=readme-ov-file#pre-processing-dataset

and what is the difference between the inpainted_1x and inpainted_1d5x

posterllava commented 3 months ago

We randomly inpainted 0.5x additional areas (other than the ground-truth box areas) to get the "inpainted_1d5x" and use it as the training set. This is to avoid over-fitting cause inpainting is not perfect (always leave some unwanted 'clue' for model). The 'inpainted_1x' is inpainted normally and used as the evaluation set.

The inpainted background image and saliency map will some be released, too.

trouble-maker007 commented 3 months ago

@posterllava thanks, got it

zimenglan-sysu-512 commented 3 months ago

any update？will release the tools and scripts that generate the inpainted results?

posterllava commented 3 months ago

any update？will release the tools and scripts that generate the inpainted results?

We here mention again the methods used for inpainting and saliency detection in our paper, which are lama https://github.com/saic-mdal/lama and basenet https://github.com/xuebinqin/BASNet.

Hope the official script can solve your problem!