tensorflow / tensorflow/models

Localization & objectness loss weighting in frcnn

Open
#8,545 0 comments 0 reactions 3 assignees View on GitHub

@pkulzc is already working on this.

Since May 21, 2020.

models:research:odapi type:support
Dominant language
Python
Stars
77.7k
Forks
44.8k
No merged PRs in 30d

Description

https://github.com/tensorflow/models/blob/9edda005848f4dead5338b5d2d93a1dc12e6c57e/research/object_detection/meta_architectures/faster_rcnn_meta_arch.py#L2320

objectness loss is computed with batch_sampled_indices as weights which doesn't seem to include groundtruth weights, while localization loss is computed sampled_reg_indices which is a batch_sampled_indices times batch_reg_weights.

Maybe I misread the code, but in the case when groundtruth weights are provided, the weights seem to be inconsistent.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.