Data source
Data from CUHK Multimedia Lab, delivered clean.
1 dataset pulled from CUHK Multimedia Lab's releases, checked field by field and shipped the way you want them — daily, weekly, or hourly, your call.
- 1 dataset
- 1 industry
- Real rows on request
What Datadory delivers from CUHK Multimedia Lab
1DeepFashion Dataset (CUHK MMLab)
Pick a catch, see the rows.
Name any CUHK Multimedia Lab dataset and we send real rows from it — not a screenshot of rows. 1,744 datasets. Pick your catch.
Get a sampleAPI, files, or your warehouse. Daily, weekly, or hourly.
Straight answers about CUHK Multimedia Lab data
How many images are in the DeepFashion corpus?
More than 800,000 fashion images in total. The Category and Attribute Prediction subset alone holds 289,222 clothes images across 50 categories, while the In-shop subset contributes 52,712 images covering 7,982 distinct clothing items.
How far back does DeepFashion data go?
The core benchmarks were released in 2016-2017, with successive extensions layered on top: Fashion Image Synthesis in October 2017, the MMFashion toolbox in 2019, parsing masks and dense pose in 2020, and DeepFashion-MultiModal in June 2022. Every Datadory delivery is labelled with the benchmark version its annotations come from.
What annotations come with each image?
A category label from the 50-category scheme, multi-label attributes drawn from 1,000 descriptors, bounding-box coordinates, clothing landmarks with visibility flags, a train/val/test partition assignment, and a pair_id linking cross-pose and cross-domain shots of the same garment. Parsing masks, dense pose and textual descriptions cover the extended benchmarks.
Can DeepFashion support visual-search prototyping?
Yes — that is what the retrieval benchmarks are built for. The In-shop subset pairs roughly 200,000 cross-pose and cross-scale images across 7,982 items, giving supervised ground truth for same-garment-new-angle matching, and pair_id makes evaluation scoring mechanical rather than manual.