search

dbolya / yolact

A simple, fully convolutional model for real-time instance segmentation.
MIT License
4.98k stars 1.32k forks source link

seperate inference yolact and yolactplus #269

Open abhigoku10 opened 4 years ago

abhigoku10 commented 4 years ago

@dbolya great work in yolact plus , read the paper have done a amazing work . Having few queries

  1. do we have separate training and inference pipeline for yolact and yolact_plus ? 2.I wanted to replace the FPN with bi-FPN which section of the code can i perform that
dbolya commented 4 years ago
  1. YOLACT++ and YOLACT share their training and inference pipeline. You can think of YOLACT++ just YOLACT with some extra settings turned on.
  2. If you wanted to modify the FPN, the FPN class is in yolact.py.
abhigoku10 commented 4 years ago

@dbolya thanks for update , Pt2. i have modified the FPN to Bi-FPN but having errors like "RunTimeError: the size of tensor(a) should match the size of tensor (b) at a non singleton dimension3 " , i am stuck in solving this issue donot knw where i am goign wrong

dbolya commented 4 years ago

For errors like those, go to the line that's throwing the error and print the shape of all tensors involved in the operation. You should see a size mismatch somewhere, which should help you debug that error.