@victor.baxan

photoMD · tier C

baxan_v. · 589 followers · 2026-06-16 13:16 · 4 images

eagle · fetched 2026-07-26 08:28 · 442.3 KB downloaded · 4 images

plays

1,001

likes

comments

Topic

Fixed taxonomy enforced as a schema enum. Prose instructions did not hold, and label drift makes week-over-week comparison meaningless.

Not tagged yet; the sweep is still running.

Caption

Author-written, as posted.

No caption. True for 83.6% of the corpus.

Speech (ASR)

faster-whisper large-v3-turbo behind Silero VAD, then a repetition filter.

Photo post: no audio track exists, so ASR never runs. OCR is the only channel.

On-screen text (OCR)

easyocr

EasyOCR for Latin script, macOS Vision for Cyrillic-suspect frames.

1 5J (erakl

rtx4090-24-02 · 11 chars · 2 frames