Description
The INRIA Person Dataset (Dalal and Triggs) is the canonical pedestrian-detection benchmark released alongside the original Histograms of Oriented Gradients (HOG) paper. It contains positive and negative crops with bounding-box annotations, with positive pedestrian crops normalised to 64x128 pixels, and is a foundational dataset in classical pedestrian detection.
Modality / size
- Modality: still RGB images with bounding-box annotations.
- Subjects / scenarios: pedestrians in varied outdoor scenes.
- Labels: bounding boxes; positive crops resized to 64x128 pixels.
Used by (papers)
- Cited next to
Caltech Pedestrianin the vault by detector pre-training studies. - Used as the seed dataset that newer pedestrian datasets are evaluated alongside.
Notes
- Public dataset; original release at the INRIA Lear / now Thoth team page.