FluxML/FastAI.jl

Benchmark image pipelines against ffcv

Open

#236 opened on Jun 10, 2022

View on GitHub
 (0 comments) (1 reaction) (0 assignees)Julia (592 stars) (49 forks)batch import
help wantedperformance

Description

[https://github.com/libffcv/ffcv] seems to be the state-of-the-art when it comes to fast image pipelines for training deep learning models on computer vision. Would be interesting to benchmark against this and see how we compare.

Results could motivate work on #235.

Contributor guide

Tech stack
None
Domain
machine learningperformance
Issue type
performance
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.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
needs investigation
Prerequisites
Julia programmingFlux.jl familiarityBenchmarking concepts
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
30
Research direction
The issue proposes benchmarking FastAI.jl image pipelines against ffcv. First, understand the current image pipeline implementation in FastAI.jl (likely in src/). Then, replicate ffcv's optimized pipeline in Julia or compare using similar datasets. The results could inform improvements mentioned in issue #235. Look at ffcv's documentation and performance benchmarks to design a fair comparison.