STVIR/pysot

Save cfg.TRAIN.START_EPOCH to state dict

Open

#397 opened on Jul 21, 2020

View on GitHub
 (5 comments) (0 reactions) (0 assignees)Python (4,314 stars) (1,153 forks)batch import
buggood first issuehas workaroundsmall

Description

Well #77 didn't work for me while resuming from checkpoint_18.pth. The problem is when we resume, the model and optimizer passed in the restore_from function are suitable for epoch less than 10 (till backbone is not training) because the cfg.TRAIN.START_EPOCH is 0 (passed in build_opt_lr function just before restore_from) initially so this mismatches the optimizer after backbone start training. So to resume my training , I pass the cfg.TRAIN.START_EPOCH as 19 and when build_opt_lr function receives epoch greater than 10 (i.e Backbone training starts) it produces the model and optimizer suitable for resuming. And i can resume my training.

Originally posted by @PhenomenalOnee in https://github.com/STVIR/pysot/issues/92#issuecomment-651571350

Contributor guide

Tech stack
pythonpytorch
Domain
machine learningai
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
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.
mostly clear
Prerequisites
basic understanding of PyTorch trainingfamiliarity with checkpointing
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40
Research direction
Investigate how TRAIN.START EPOCH is used in the training scripts and optimizer initialization. The fix involves saving this config value in the checkpoint state dict and restoring it during resume. Look at files like train.py or core/train.py and the restore from function. See issue #92 for additional context.
Save cfg.TRAIN.START_EPOCH to state dict · STVIR/pysot#397 | Good First Issue