yangxue0827/FPN_Tensorflow

The problem for not fixing BN

Open

#21 opened on Jun 6, 2018

View on GitHub
 (6 comments) (0 reactions) (0 assignees)Python (413 stars) (150 forks)batch import
good first issue

Description

What is the Map in pascal voc 07 dataset using this code? As i read your code, i find that you do not fix the BN parameters. The batchsize is small, following the paper FPN or Mask RCNN, you need to set the training and trainbale to False for the slim.batch_norm op. I am testing whether this is the problem for the bad generalization for mutiple objects.

Contributor guide

Tech stack
pythontensorflow
Domain
machine learning
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
4
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
half day
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
needs investigation
Prerequisites
TensorFlow basicsBatch normalization understandingObject detection with FPN
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
10
Research direction
Investigate the handling of batch normalization in the training code. Look at the model definition and training script to see if batch norm parameters are frozen. Check comments in the issue for discussion about small batch size and generalization. Possibly modify the training to set slim.batch norm training parameter to False as suggested.