Agri-CV-Research

Agri-CV Research is a Python framework designed to automate the full process of conducting a computer vision research project in the field of agriculture. The idea is that a researcher can start from a topic like "detecting tomato leaf disease with a lightweight model" and the framework will guide or automate each stage: searching related papers, generating research ideas, designing experiments, training models, evaluating results, creating publication figures, and drafting a paper. All nine stages are meant to run in sequence from a single command-line interface.

The agricultural focus means the framework ships with support for problems specific to farming and crops: identifying plant diseases, recognizing weeds in fields, counting fruit in orchards, and monitoring crop growth. It includes connectors for several well-known public datasets (PlantVillage, PlantDoc, DeepWeeds, MinneApple) and wrappers for popular model families including YOLOv8 for detection, Vision Transformers and EfficientNet for classification, SAM for segmentation, and CLIP for zero-shot or few-shot tasks.

Evaluation goes beyond standard accuracy numbers. The framework tests model robustness under conditions that appear in real agricultural deployment: rain, fog, low light, and partial obstructions. It also includes edge-device benchmarking, which matters when models need to run on hardware in the field rather than in a data center.

On the output side, the framework generates figures and tables formatted for research papers: Grad-CAM attention maps, confusion matrices, training curves, and LaTeX-ready artifacts that can go directly into a manuscript. There is also a paper-writing command that takes experiment results and produces a structured draft.

The project is designed to work alongside AI coding assistants such as Cursor and Claude Code, and the README describes specific roles for each. Installation requires Python 3.10 or later and is done through pip. The repository is released under an MIT license.