Precision at Scale: Domain-Specific Datasets On-Demand
cs.CV, cs.AI
Submitted: 2024-07-03
Updated: 2026-09-05
Comments: Accepted to Pattern Recognition, vol. 171 (2026), art. 112236. Code: https://github.com/jesusmolrdv/Precision-at-Scale Data & Models: https://huggingface.co/jesusmolrdv
Journal ref: Pattern Recognition, 171 (2026) 112236
DOI: 10.1016/j.patcog.2025.112236
Code: https://github.com/jesusmolrdv/Precision-at-Scale
License: http://creativecommons.org/licenses/by-nc-nd/4.0/
The gist: In the realm of self-supervised learning (SSL), conventional wisdom has gravitated towards the utility of massive, general domain datasets for pretraining robust backbones.
Terminology
Abstract
In the realm of self-supervised learning (SSL), conventional wisdom has gravitated towards the utility of massive, general domain datasets for pretraining robust backbones. In this paper, we challenge this idea by exploring if it is possible to bridge the scale between general-domain datasets and (traditionally smaller) domain-specific datasets to reduce the current performance gap. More specifically, we propose Precision at Scale (PaS), a novel method for the autonomous creation of domain-specific datasets on-demand. The modularity of the PaS pipeline enables leveraging state-of-the-art foundational and generative models to create a collection of images of any given size belonging to any given domain with minimal human intervention. Extensive analysis in two complex domains, proves the superiority of PaS datasets over existing traditional domain-specific datasets in terms of diversity, scale, and effectiveness in training visual transformers and convolutional neural networks. Most notably, we prove that automatically generated domain-specific datasets lead to better pretraining than large-scale supervised datasets such as ImageNet-1k and ImageNet-21k. Concretely, models trained on domain-specific datasets constructed by PaS pipeline, beat ImageNet-1k pretrained backbones by at least 12% in all the considered domains and classification tasks and lead to better food domain performance than supervised ImageNet-21k pretrain while being 12 times smaller. Code repository: https://github.com/jesusmolrdv/Precision-at-Scale/
Sources
- Foundational Models Defining a New Era in Vision: A Survey and Outlook
- Synthetic Data from Diffusion Models Improves ImageNet Classification
- BEiT: BERT Pre-Training of Image Transformers
- Muse: Text-To-Image Generation via Masked Generative Transformers
- MMDetection: Open MMLab Detection Toolbox and Benchmark
- SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?
- Mixtral of Experts
- FoodX-251: A Dataset for Fine-grained Food Classification
- Segment Anything
- BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
- UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
- DINOv2: Learning Robust Visual Features without Supervision
- How to train your ViT? Data, Augmentation, and Regularization in Vision Transformers
- Learning Vision from Models Rivals Learning Vision from Data
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- SimVLM: Simple Visual Language Model Pretraining with Weak Supervision
- Demystifying CLIP Data
- CoCa: Contrastive Captioners are Image-Text Foundation Models
- Oktoberfest Food Dataset
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models