search

JianqiangWan / Super-BPD

Super-BPD: Super Boundary-to-Pixel Direction for Fast Image Segmentation (CVPR 2020)
Apache License 2.0
200 stars 37 forks source link

How to test a single RGB image? #4

Closed carfei closed 4 years ago

carfei commented 4 years ago

Thank you very much for your work. If I want to test an RGB image, can I get a semantic segmentation graph through the model input picture, and I want clear visualization results, can I get a more precise semantic segmentation graph?

JianqiangWan commented 4 years ago
  1. Compile and install the post_process program.
  2. Transform input image to a two channel direction map (pred_flux), refer to test.py.
  3. https://github.com/JianqiangWan/Super-BPD/blob/3c44638117625001ea0a92616f339c6ae4d5b956/post_process/demo.py#L34-L42
  4. The final segmentation result has no semantic information，and post-processing includes the construction and connection process of the Region Adjacency Graph (RAG), we didn't output the RAG.