marco-garosi

marco-garosi / CIRCLE

Public

[CVPR Findings 2026] Large Multimodal Models as General In-Context Classifiers

circle-lmm.github.io
19
0
100% credibility
Found Mar 07, 2026 at 19 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

CIRCLE is a research framework for testing large AI models that handle images and text on classification tasks, using few examples to teach them without extra training.

How It Works

1
🔍 Discover CIRCLE

You stumble upon this clever tool while exploring ways AI can classify images using just a handful of examples, like teaching it on the fly.

2
📥 Set it up

Download the tool and prepare your computer so everything is ready to test AI models without any hassle.

3
🤖 Pick AI helpers

Choose smart AI models that understand pictures and words, deciding which ones to put through their paces.

4
🖼️ Select picture challenges

Pick collections of images and categories to test how well the AIs guess what's in new photos.

5
🚀 Launch the tests

Hit start and feel the excitement as the tool runs trials, letting AIs learn from examples to classify unknowns.

6
📊 Check the scores

Review clear reports showing how accurate each AI is, with numbers on matches and smart comparisons.

🏆 Spot the winners

Celebrate finding top-performing AIs with handy rankings, ready to share insights on the best classifiers.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 19 to 19 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is CIRCLE?

CIRCLE is a Python benchmarking tool for turning large multimodal models (LMMs) into general-purpose in-context classifiers on image datasets like Caltech101, Food101, and Flowers102. From the CVPR Findings 2026 track—part of the buzz around CVPR 2026 GitHub repos and CVPR findings workshop—it evaluates zero-shot and few-shot performance against CLIP-style VLMs, while introducing a training-free CIRCLE method for open-world classification via iterative pseudo-label refinement. Developers get CLI scripts to run evals on models like Qwen2-VL-7B, compute metrics offline, and generate Elo rankings, all with Slurm or local batch support.

Why is it gaining traction?

It stands out with dead-simple scheduling via bash scripts for multi-GPU local runs or distributed Slurm jobs, plus offline metric computation to avoid wasting compute on heavy model-based evals like textual inclusion. FlashAttention integration speeds up inference, and utilities for cloning configs make experimenting with variants effortless—no more manual YAML tweaks. In the circle community GitHub space, it's a practical drop-in for CVPR 2024 papers GitHub and CVPR 2025 papers GitHub workflows, bridging LMM hype to reproducible benchmarks.

Who should use this?

Computer vision researchers benchmarking LMMs against VLMs on closed- and open-world classification. ML engineers at circle internet group-style teams evaluating Qwen-VL or LLaVA for production classifiers. Academic labs running CVPR findings 2026 baselines on standard datasets without rebuilding eval harnesses from scratch.

Verdict

Grab it if you're deep into LMM classification evals—solid docs, pre-commit hygiene, and Slurm-ready scripts make it production-usable out of the box. With just 19 stars and a 1.0% credibility score, it's early-stage and low-adoption, so pair it with mature alternatives like lmms-eval for safety, but its CVPR Findings 2026 pedigree signals strong potential.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.