The audio extract length is meant for visualizing/grabbing sequences that can
run minutes long, but the control capped it and stepped in fiddly 0.10s
increments. Raise the range to effectively unlimited (24h; ffmpeg stops cleanly
at end-of-file if the source is shorter) and make the arrows step 1s — typing
still allows sub-second precision. Widen the field for the larger values.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A dedicated "♪ Extract audio" button on the transport row grabs an exact
length of audio (set via the adjacent length box, from the playhead) and opens
a Save As dialog. Output format follows the chosen extension — WAV (pcm_s16le),
MP3 (libmp3lame), FLAC, m4a/aac, ogg/opus — re-encoding as needed; unknown
extensions let ffmpeg pick from the container.
- core.ffmpeg.build_audio_clip_command(input, start, duration, out_path):
fast-seek + exact -t duration + -vn, codec by extension. Verified end-to-end
(wav/mp3/flac all land at exactly the requested duration).
- Timeline shows the audio area as a distinct teal dashed band spanning
[cursor, cursor+length], updated live as the playhead or length changes, so
you see exactly what will be extracted.
- Length + last save dir persist in QSettings; button enabled once a file loads.
Tests: 3 core (codec-by-extension, exact length, case-insensitive) + 2 GUI
(controls exist, band tracks cursor/length).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two issues with the per-subprofile (subcategory) export buttons:
1. Visibility was decided by a fuzzy `f.endswith("_" + suffix)` match against
the hidden-subcats set. A ghost "_blowjob" (empty-base leftover from the
trailing-slash folder bug) or an unrelated "mp4_no_clap" would match and
hide the wrong button — so enabling a subcategory in the Sub menu never
revealed its export button. Match the exact "<base>_<suffix>" folder name
instead (same name the menu shows and _hidden_subcats stores).
2. The buttons were crammed into the transport row after Export. Move them to
their own row with stretches on both ends so the (often many) "▸ name"
buttons stay centered and out of the transport controls.
Also cleared the polluted hidden_subcats/POV_Front set in the user's QSettings
(ghost "_*" names + a hide-all'd set of real "mp4_*"), so every subcategory is
visible again. Regression test added for the exact-match predicate.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
A folder ending in "/" made os.path.basename() return "", so subprofile
folders/labels became "_blowjob" instead of "mp4_blowjob" — cluttering the
subcategory menu and breaking the marker↔category match. rstrip the trailing
separator in _tab_export_folder and the three basename(_txt_folder) sites.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
".../AlexisCrystal/" + "_copy" was producing ".../AlexisCrystal/_copy"; rstrip
the trailing separator first → ".../AlexisCrystal_copy". Regression test uses a
trailing-slash source folder.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Constructing MainWindow loads and (on close) re-saves the playlist tabs; a test
that mutated tab state could persist into the user's real session. Redirect
QSettings to a temp dir at import time.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New tab copies the source tab's video list + separators, gets a unique
"<name> copy" label and an adapted own export folder ("<folder>_copy"), and
inherits the tab-named-folder flag. No files are moved or copied — you export
into the new tab's folder. Keeps Foley/variant datasets separate without the
file-shuffling that a misexport used to require.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Pinning 2 of 3 panels previously showed a 3rd "leftover" tab-column, which
read as all-three-pinned and was confusing. Now the split view shows exactly
the pinned panels (pin 2 -> 2 columns, pin 3 -> 3). Adds an always-available
View > Side-by-side panels submenu of checkable toggles as the way to pin a
panel while already in split view (the right-click-tab gesture only works in
tabbed mode). Tests assert exactly-N-columns and the menu-pin path; the win
fixture now resets deck state so tests don't depend on persisted layout.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- test_deck_stack_exists: _deck_stack present; default shows _control_deck.
- test_pinning_two_panels_switches_to_split: pin 2 panels + refresh →
stack shows _deck_split_container.
Pin via _pinned flags directly (not the toggle handler) so no QSettings
write leaks into other function-scoped windows; existing 6 tests run in
default/tabbed state and still pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- Export layout changed from clip_NNN group dirs to vid_NNN per-video folders
- Automatic DB migration rewrites old paths and moves files on startup
- Per-video counter with DB cross-check to prevent overwrites
- Changelog popup on version bump with "don't show again" checkbox
- Scan region resize now requires Shift+drag to prevent accidental edits
- Recalculate vid folder and counter on file load
- Add EAT_LARGE embedding model variant
- Update tests for new flat export path structure
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Use microsecond-precision timestamps to prevent version merging on
sub-second scans
- Clear undo stack when switching scan versions (stale row references)
- Parse timestamp labels robustly instead of hard-coded string slicing
- "Clear All" in hard negatives dialog respects active model filter
- Remove time.sleep from tests (no longer needed with microsecond timestamps)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add source_model column to hard_negatives table with migration. New
get_hard_negatives() returns full rows, delete_hard_negatives_by_ids()
for bulk deletion. get_training_data() gains use_hard_negatives param.
TrainDialog has "Use hard negatives" checkbox. Scan panel passes current
model name when marking negatives.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add scan_timestamp column to scan_results. save_scan_results now inserts
with a timestamp and prunes versions beyond max_versions (default 5).
get_scan_results returns only the latest version by default, with optional
scan_timestamp parameter for loading specific versions. New get_scan_versions
method returns available versions for a (file, profile, model) tuple.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Ghost folders (scan-export-only) no longer appear in training dropdowns.
Also filters out 0-clip folders from get_training_stats.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix test_utils.py importing build_annotation_json_path from main
instead of core.annotations (all 59 tests pass now)
- Fix get_training_data double-counting clips at same start_time
in both positive and soft sets — subtract positive from soft
- Add cancel_flag to train_classifier so training can be interrupted
between videos (TrainWorker passes self as cancel_flag)
- Remove orphaned core/export.py (was for deleted server API)
- Remove stale Dockerfile and docker-compose.yml (referenced server)
- Clean up leftover server/__pycache__ and client/ build artifacts
- Add torch to requirements.txt (was only mentioned in comments)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove legacy distance-mode scanning (build_profile, _similarity, etc.)
and hand-crafted intensity features — pipeline is now embedding-only
- Integrate Microsoft BEATs as embedding option alongside wav2vec2/HuBERT
- Add TrainDialog with positive class selector, model picker, video dir
fallback, and live training stats
- Add TrainWorker QThread with cancel support and proper lifecycle cleanup
- Add source_path column to DB for robust source video tracking
- Add get_export_folders/get_training_data/get_training_stats to DB
- Wire source_path in all export DB writes (_on_clip_done, _on_auto_clip_done)
- Cancel scan/train workers in closeEvent to prevent use-after-free crashes
- Add setup_env.sh supporting both conda and python venv (CUDA 12.8)
- Update requirements.txt with all actual dependencies
- Update 8cut_train.py with --positive flag for new DB-driven training
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Root cause of poor discrimination: MFCC[0] (energy) dominated the
feature vector, making cosine similarity see all audio as similar.
Changes:
- Skip MFCC[0], use 12 coefficients instead of 20
- Add delta MFCCs for temporal dynamics
- Add 7-band spectral contrast for tonal vs noise quality
- Switch from cosine similarity to euclidean-distance-based score
- Pre-compute STFT once for whole file (10-20x faster)
- Vectorized sliding window via cumulative sums (no Python loop)
- Lower sample rate 22050→16000 Hz (faster, no quality loss)
- 62-dim feature vector (was 40-dim mean+std of raw MFCCs)
- Default threshold 0.05 (new similarity scale)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Mean-only vectors were too similar across different audio segments,
causing everything to match even at threshold 0.99. Adding std
captures temporal dynamics and makes the similarity scores much
more spread out.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
EthanfelandClaude Opus 4.8