Embrapa Wine Grape Instance Segmentation Dataset - Embrapa WGISD
Embrapa WGISD (Wine Grape Instance Segmentation Dataset) was created to provide images and annotation to study object detection and instance segmentation in image-based monitoring and field robotics for viticulture. It provides instances from five different grape varieties taken on field. These instances shows variance in grape pose, illumination and focus, including genetic and phenological variations as shape, color and compactness.
Possible uses include relaxations of the instance segmentation problem: classification (Is a grape in the image?), semantic segmentation (What are the “grape pixels” in the image?), object detection (Where are the grapes in the image?). The WGISD can be also in grape variety identification.
The building of the WGISD dataset was supported by the Embrapa SEG Project 01.14.09.001.05.04, Image-based metrology for Precision Agriculture and Phenotyping, and the CNPq PIBIC Program.
Each instance consists in a RGB image and an annotation describing grape bunches locations as bounding boxes. A subset of the instances also contains binary masks identifying the pixels belonging to each grape bunch. Each image presents at least one grape bunch. Some grape bunches can appear far at the background and should be ignored.
File names prefixes identify the variety observed in the instance:
| Prefix | Variety |
|---|---|
| CDY | Chardonnay |
| CFR | Cabernet Franc |
| CSV | Cabernet Sauvignon |
| SVB | Sauvignon Blanc |
| SYH | Syrah |
The dataset consists of 300 images containing 4,398 grape bunches identified by bounding boxes. A subset of 125 images also contains binary masks identifying the pixels of each bunch. That means that from the 4,398 bunches, 1,898 of them presents binary masks for instance segmentation.
| Prefix | Variety | Date | Instances | Boxed bunches | Masked bunches |
|---|---|---|---|---|---|
| CDY | Chardonnay | 2018-04-27 | 65 | 838 | 242 |
| CFR | Cabernet Franc | 2018-04-27 | 65 | 1048 | 460 |
| CSV | Cabernet Sauvignon | 2018-04-27 | 57 | 640 | 290 |
| SVB | Sauvignon Blanc | 2018-04-27 | 65 | 1313 | 586 |
| SYH | Syrah | 2017-04-27 | 48 | 559 | 256 |
| Total | 300 | 4398 | 1834 |
Each instance contains a 2048 x 1365 pixels RGB image and a text file containing one bounding box description per line. These text files follows the “YOLO format”:
class cx cy w h
class is an integer defining the object class - the dataset presents only the grape class that is numbered 0, so every line starts with this “class zero” indicator. The center of the bounding box is the point (cx, cy), represented as float values because this format normalize the coordinates by the image dimensions. To get the absolute position, use (2048 * cx, 1365 * cy). The bounding box dimensions are given by w and h, also normalized by the image size.
The instances presenting mask data for instance segmentation contain files presenting the .npz extension. These files are compressed archives for Numpy n-dimensional arrays. Each array is a 1,365 x 2,048 x n_bunches three-dimensional array where n_bunches is the number of grape bunches observed in the image. After assigning the Numpy array to a variable M, the mask for the i-th grape bunch can be found in M[:,:,i]. The i-th mask corresponds to the i-th line in the bounding boxes file.
The dataset also includes the original image files, presenting resolutions bigger than 2048 x 1365 pixels. The normalized annotation for bounding boxes allows easy identification of bunches in the original images, but the mask data will need proper rescaling by users that wish work in the original full resolution.
Everything is included in the dataset.
The dataset comes with specified train/test splits. The splits are found in lists stored as text files. There are also lists referring only to instances presenting binary masks.
| Images | Bounding boxes | Masks | |
|---|---|---|---|
| Train/Val | 240 | 3555 | 1402 |
| Test | 60 | 843 | 432 |
| Total | 300 | 4398 | 1834 |
- Precision/recall
- AP
- AP50