Chihiro-n

Chihiro-n / csiro-biomass-agentic-solution

Public

5th place solution repository for Kaggle CSIRO Image2Biomass, including the experiment workflow used with Claude Code and Codex.

18
1
100% credibility
Found Mar 17, 2026 at 18 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Shared code and experiment logs for a top-ranked solution predicting pasture biomass components from aerial images in a public contest.

How It Works

1
🔍 Discover the winning plant predictor

You stumble upon this shared secret recipe from a top contest where photos of grass fields turn into weight estimates, created mostly by smart helpers.

2
📥 Gather your field photos

You grab the collection of real pasture pictures with known weights from the contest site to get started.

3
⭐ Choose your setup

You pick one of the ready-made plans that worked best, like a super-detailed viewer for tiny plant details.

4
🧑‍🔬 Let it learn patterns

You run the simple guide on a cloud playground, feeding it example photos so it learns to spot green, dead, and clover amounts.

5
🧪 Predict on new fields

You feed in fresh test photos from unknown pastures and watch it guess the biomass weights quickly.

🏆 Celebrate spot-on estimates

You get reliable predictions for total grass, green growth, dead bits, clover, and more, just like the gold medal winner.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 18 to 18 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 csiro-biomass-agentic-solution?

This Python repo delivers a 5th place solution (gold medal) to the Kaggle CSIRO Image2Biomass challenge, predicting pasture biomass components—green, dead, clover, GDM, and total dry mass—from aerial images. It provides Colab notebooks for training vision transformer models like DINOv3 at high resolutions (up to 960px), Kaggle-ready inference scripts with test-time training and TTA, plus YAML configs and experiment logs for quick reproduction. Users get a full agentic workflow powered by Claude Code and Codex, turning raw images into precise kg/m² estimates for agrotech applications.

Why is it gaining traction?

Unlike polished Kaggle winners, this preserves the messy, real agentic process—AI agents iterating via short prompts on configs and pipelines—showing how LLMs nailed a top-5 LB score (0.66 private) out of 3,803 teams. Developers dig the density integration heads for spatial biomass summing, gradual unfreezing, and CV strategies mimicking top discussants, plus fast FP16 inference. It's a blueprint for biomass estimation blending vision models with agentic coding.

Who should use this?

Kaggle competitors in image regression competitions, especially remote sensing or agriculture CV tasks needing multi-head biomass prediction. Agrotech engineers building pasture monitoring tools from drone imagery. Devs experimenting with Claude/Codex for agentic ML pipelines on Colab/Kaggle.

Verdict

Grab it as a strong baseline for CSIRO-style biomass challenges or agentic experiments—docs shine via writeups and EXP summaries. With 18 stars and 1.0% credibility, it's early-stage but battle-tested; validate on fresh data before production. (198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.