Update app.py

Set examples_per_page=16 (all 16 thumbnails on one page)

Point Paper badge to arXiv https://arxiv.org/pdf/2604.21681

Pose: enable all 4 sizes; default = 1B (shipped)

Pose: lock radio to 0.4B during iteration mode

Pose: per-person bbox-scaled radius+thickness; green bbox overlay

Pose: drop ImageSlider, use plain Image (skeleton already overlays input)

Pose: match seg/normal aesthetic (Helvetica, gradient title, badges, ImageSlider, hero examples, 0.4B-only preload)

Switch pose detector from mmdet/RTMDet to DETR (transformers, Apache 2.0)

Pin Python 3.10 + torch 2.1.2; vendor sapiens2 to bypass requires-python

Initial sapiens2-pose Space (HF download at startup, all 4 sizes)