The current work is published in the paper "A Comparative Analysis of Sensor-, Geometry-, and Neural-Based Methods for Food Volume Estimation" (https://doi.org/10.1145/3607828.3617794).
The dataset consists of two folders, one for the plastic and one for the real food. In every meal folder there are the following subfolders for distances 40 and 60 cm:
- The "GT_RECAP": Containing the point clouds for each food item, and the total meal.
- The "INTELRS": The original RGB image (image_1_original.jpg) and original depth image (image_1_original_depth.png) captured by the Intel RealSense D455 sensor, the segmented food items (mask.png) and the information about each segmented food item (details.txt).
- The "LIDAR": The original RGB image (image_1_original.jpg) and original depth image (image_1_original_depth.png) captured by the iPhone 14 Pro integrated with a LiDAR sensor, the scaled depth (real_depth.npy), the segmented food items (mask.png) and the information about each segmented food item (details.txt).
- The "STEREO": The original RGB images from 90 and 75 degrees (image_1_original.jpg, image_2_original.jpg) captured by the OnePlus 7 Pro, the segmented food items (mask.png, mask2.png), the gravity data (Gravity_image_1.json, Gravity_image_2.json) and the information about each segmented food item (details.txt).
- The "ZOE": The original RGB image (image_1_original.jpg) captured by the iPhone 14 Pro , the segmented food items (merged_mask.png) and the information about each segmented food item (details.txt).
Additionally, you can find the "Volume GT Meals_MADIMA2023.xlsx" that contain all the ground truth volumes and the "gocarb.jpg" image of the reference card with actual Size: 8.5cm*5.5cm.
- Downloads last month
- 0