Introduction
Multi-cultivar and multi-view rice plant image dataset (CVRP) consists of 2,203 field images of rice plants, captured from 231 landraces and 50 modern cultivars grown under dense planting conditions. Each image is paired with its corresponding annotated mask. Additionally, we provide 123 indoor images focused on individual panicles.
CVRP Overview
Annotation Workflow
A semi-automatic annotation method using deep learning models was developed for annotations, which were then subjected to strict manual curation.
Multiple Cultivars and Multiple Viewpoints
There are 8 landraces used in the study with distinctive characteristics of panicles. The landraces are: (a) Shuiniupi, (b) Luganbai, (c) Jinzhongjing, (d) Gankeqing, (e) Liushizijing, (f) Xiangfandao, (g) Hongkenuo, and (h) Hahaxiao.
Images of rice plants with panicles in the field from four distinct viewpoints: nadir view, oblique view, side view and close-up view.
Annotation Results
With the exception of a few landraces with poor growth conditions, eight images were selected for each plant, aiming to capture as many different viewpoints as possible.
Individual Panicles
We captured images of individual panicles at Nanjing Agricultural University's lab: one set in their natural state with a homemade base, preserving pose and structure; the other in a manually unfolded state to reveal hidden grains and internal structures. Examples are Jinzhongjing and Hahaxiao in both natural and unfolded states.
CVRP Structure
CVRP is organized into several directories, each representing different aspects of the rice plants and panicles. The overall structure is as follows:
/CVRP
│
├── /FieldImages
│ ├── /T1
│ ├── /Images
│ └── /Annotations
│ ├── /T2
│ ├── /Images
│ └── /Annotations
│ ├── /T3
│ ├── /Images
│ └── /Annotations
│ ...
└── /IndoorPanicleImages