GitHub - lixinustc/Enhance-Anything: Enhance and restore any things continuously with Segment Anything (original) (raw)

Purpose: We aim to achieve a unfied intellignet image/video PS/enhancement framework. More functions will be developed.

Brief intro for current function: Enhance and restore anything continuously with Segment Anything Model for image processing (The initial version is achieved with CResMD, and it will be extended in future).

🌟 Enhance Anything Features

Continuously Enhance Anything by a simple click
Assist Colorizing Anything
Assist multiple distortion types
Assist multiple enhancement types
Technique Analysis - Arxiv Version
Demo
Improving generalization capability

Visual Examples

Enhance zebra continuously

Enhance dog and chair continuously

Enhance dog continuously

Enhance butterfly continuously

Usages

Please download the weights for segment-anything, and move it to ./pretrained_weights. Then you can run the code with:

 python enhance_anything_continusly.py  --input_img ./examples/0006.png    --point_coords 200 300    --point_labels 1     --dilate_kernel_size 30     --output_dir ./results     --sam_model_type "vit_h"     --sam_ckpt ./pretrained_models/sam_vit_h_4b8939.pth --opt ./CResMD/options/test/modulation_CResMD.yml

The results are shown in ./results_continuous/.

Environment

conda create -n seganything python python -m pip install torch torchvision torchaudio python -m pip install -e segment_anything python -m pip install lmdb