``ScatSpotter'' --- A Dog Poop Detection Dataset

Jonathan P Crall

Keywords: poop, feces, dataset, dataset distribution, detection, segmentation, IPFS, BitTorrent, HuggingFace

TL;DR: We build the largest open dataset of dog poop images, train MaskRCNN and VIT models, and experiment with centralized and decentralized distribution mechanisms.

Abstract: We introduce a new dataset containing phone images of dog feces, annotated with manually drawn or AI-assisted polygon labels. Its over 9000 ``before/after/negative'' full resolution images contain 6000 polygon annotations. The collection and annotation of images started in late 2020. This paper focuses on two checkpoints from 2025-04-20 and 2024-07-03. We train VIT and MaskRCNN baseline models to explore the difficulty of the dataset. The best model achieves a pixelwise average precision of 0.858 on a 691-image validation set and 0.810 on a small independently captured 121-image contributor test set. Dataset snapshots are available through four different distribution methods: two centralized (Girder and HuggingFace) and two decentralized (IPFS and BitTorrent). We study of the trade-offs between distribution methods and discuss the feasibility of each with respect to reliably sharing open scientific data. The code for experiments is hosted on GitHub. The data license is CC-BY 4.0. Model weights are available with the dataset. Experiment hardware, time, energy, and emissions are quantified.

Dataset URL: https://huggingface.co/datasets/erotemic/scatspotter

Code URL: https://github.com/Erotemic/scatspotter

Primary Area: Datasets & Benchmarks for applications in computer vision

Submission Number: 205