Prize Money via AI Institute for Food Systems (AIFS)
1st prize: $2500, 2nd prize: $1000, 3rd prize: $500 (for each challenge)
AgML Crop Detection Generalizability Challenge
Participants will be provided 6 to 8 image-based crop detection datasets from AgML for model training, with an additional 4 to 6 datasets held out for model evaluation (~10,000 images total). Each dataset will be contain a different fruit, vegetable, or nut imaged in an agricultural environment using various camera types, viewpoints, and lighting conditions. The goal of this challenge is create a crop detection model that generalizes well to out-of-distribution crop types and imaging conditions. A “generalizability” evaluation metric will be defined as the mean average precision @ [0.5 : 0.95] averaged across all test datasets.
AgML Syn2Real Crop Detection Data Efficiency Challenge
Participants will be provided synthetic crop detection image datasets for model training (via AgML’s Helios synthetic data generation API), and a small subset of real world data from the same crop type. The objective for this challenge is to train a high performing real crop object detection model while fixing the number of real images available during training (i.e. 5, 10, or 20). A ”data efficiency” evaluation metric will be defined as the mean average precision @ [0.5 : 0.95] on a held out set of real crop images averaged across models trained with a fixed set of 5, 10, or 20 real images.