open-mmlab/mmrotate

[Feature] WANDB validation metrics

Open

#828 opened on Apr 24, 2023

View on GitHub
 (1 comment) (0 reactions) (2 assignees)Python (641 forks)github user discovery
good first issue

Repository metrics

Stars
 (2,141 stars)
PR merge metrics
 (No merged PRs in 30d)

Description

What's the feature?

I'm running training sessions based on the REDET base config with an added WANDB hook:

base_cfg.log_config.hooks = [
    dict(type='TextLoggerHook'),
    dict(type='TensorboardLoggerHook'),
    dict(type='MMDetWandbHook',
         init_kwargs=train_cfg['wandb_config'],
         interval=10,
         log_checkpoint=False,
         log_checkpoint_metadata=False,
         num_eval_images=100,
         bbox_score_thr=0.3)
]

Currently the WANDB hook only reports mAP for validation epochs image

I need to track recall as well, which is calculated and available in the training log, but not in WANDB. Ideally, I'd like to also see recall and precision for various classes and object sizes.

Am I missing something about the hook? Do I need to create a new hook? If I do, how do I access all available metrics?

Any other context?

No response

Contributor guide

Research direction
Investigate how to extend the MMDetWandbHook to log additional metrics like recall and precision. Look at the hook's source code in mmdetection and check how validation metrics are accessed. Consider subclassing the hook or modifying the evaluation pipeline to log these metrics.
Tech stack
pythonpytorch
Domain
machine learningai
Issue type
Feature
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.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Mostly clear
Prerequisites
PythonPyTorchmmdetectionmmrotate
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
40