Video Processor using dghs-imgutils

Upload one or more videos, or a sequence of images. Videos are processed individually, while multiple images are treated as a single sequence. Each processed source (video or image sequence) is then sequentially analyzed by dghs-imgutils to detect subjects, classify items, and process its content according to your settings, ultimately generating a ZIP file with the extracted images.

Detection Flow: Person ⇒ Half-Body (if no person) ⇒ Face (on target) ⇒ Head (on target).

Analysis Flow: CCIP Clustering ⇒ LPIPS Clustering ⇒ Aesthetic Labeling.

Note on CCIP Folders: CCIP cluster folders are named {source_prefix}_ccip_XXX, sorted by image count (most images = _ccip_000).

Note on LPIPS Folders: LPIPS cluster folders (e.g., lpips_XXX or lpips_sub_XXX) are also sorted by image count within their scope. 'noise' folders are named explicitly.

Detection Options
0 1
PD Model
0 1
0 1

Detects half-bodies in full items if Person Detection is off/fails.
0 1
HBD Model
0 1
0 1
0 1
FD Model
0 1
0 1

Detects heads in targets. Crops if meets width req.
0 1
HD Model
0 1
0 1

Analysis & Classification

Clusters results by similarity. Folders sorted by image count.
CCIP Model
0 1

Clusters images by LPIPS similarity. Applied after CCIP (if enabled) or globally. Folders sorted by image count.
0 1

Prepends aesthetic label to filenames.
Aesthetic Model

Warning: Complex combinations can be slow. Models downloaded on first use.