Fetching metadata from the HF Docker repository...
Jun 25 10:30 Update app.py from main Jun 9 17:53 Enlarge tab caption font size usamaalam +1: Claude Opus 4.8 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> 17:49 Switch to complete dark professional theme usamaalam +1: Claude Opus 4.8 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> 17:43 Fix header/footer consistency — full-bleed banner, unified page background usamaalam +1: Claude Sonnet 4.6 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> 17:38 Add Model & Training tab with professional header/footer usamaalam +1: Claude Sonnet 4.6 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> 16:37 Add CASIA scope note to UI 13:38 Add debug instrumentation (build version, input order, tensor stats, raw pred) 11:56 Fix ELA preprocessing to match training notebook (amplify + JPEG + tf.image decode) 10:08 Fix ELA computation to match train.py (ImageChops.multiply, not Brightness) The model trained on attenuated/dark ELA images (multiply divides by 255),
but app.py was amplifying them via Brightness.enhance, causing every image
to saturate the ELA branch and predict FORGED. 09:51 Reference M3_best_v2.h5 (full model saved from TF 2.20) 09:48 Upgrade to Streamlit 1.40 + TF 2.20 (resolves protobuf conflict); fix use_container_width 09:34 Load full H5 model via load_model; pin TF 2.20 to match training runtime 04:48 Fix build_model to exactly match train.py architecture and preprocessing 04:19 Fix missing layers import in build_model 04:15 Clean deployment: model loaded from HF Hub at runtime usamaalam +1: Claude Sonnet 4.6 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> 
