1 of 5

Introduction

Project Title:

AgriCLIP – Adapting CLIP for Agriculture and Livestock

Team Introduction:

One-liner about the research

paper:

AgriCLIP is a vision-language foundational model designed to overcome domain gaps in agriculture and livestock by using a specialized dataset and training pipeline.

2 of 5

Understanding of Base paper

Problem Solved:

Lack of large-scale image-text datasets in agriculture.

Existing CLIP performs poorly on fine-grained agricultural tasks.

Key Idea: AgriCLIP improves CLIP’s performance in agriculture by addressing domain gaps.

Core Problem:

General CLIP models are trained on everyday images, not agricultural ones. This creates a gap that makes them perform poorly on agriculture-related tasks

3 of 5

Problem Defination

Challenges in Agriculture:

Most datasets are task-specific (e.g., only disease classification).

Lack of paired image-text data for crops, livestock, and fishery.

Scope of project

In-Scope: ALive dataset creation (600K image-text pairs). Zero-shot classification tasks.

​

Out-of-Scope: Segmentation, object detection, and real-time deployment.

4 of 5

Architechtural Diagram

5 of 5

Conclusion & Future Horizons

​

Impressive Results:

Evaluated on 20 downstream agricultural datasets (over 300K images), AgriCLIP achieved an average accuracy of 48.27%, demonstrating a significant +9.07% improvement over standard CLIP models.

  • AgriCLIP bridges the domain gap in agriculture & livestock.

​

  • Combines ALive dataset (600K image-text pairs) with a hybrid training pipeline.

​

  • Learns both semantic features (global meaning) and fine-grained features (disease spots,
  • breed differences).