Replace fine thumbnail grid with mpv/vlc/ffplay handoff

Watching the video play turns out to be much faster than scanning a
thumbnail grid. The coarse 10-min thumbnail grid still does rough
localisation; after picking, a video player launches at coarse_t-30s
paused with frame-accurate scrubbing controls. The analyst reads the
exact opening time off the player's OSD and types it into the
terminal prompt (default = the coarse pick, so a single Enter keeps
the coarse pick if the player is hard to use).

Backend auto-detects mpv > vlc > ffplay; gracefully degrades to "use
the coarse pick" if no player is installed.

New `bad_rois` column captures non-opening sub-arenas (partial-opening
videos like the 2024-10-21 set where only the lower half opens). The
prompt validates entries are integers in 1..6.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Giorgio Gilestro 2026-05-01 12:20:09 +01:00
parent 125f187187
commit 2b75daa783

View file

@ -1,25 +1,25 @@
"""Interactive picker for barrier-opening time, frame-by-frame thumbnail style.
For each video that doesn't yet have a barrier-opening annotation, show a
10x6 grid of timestamped thumbnails extracted directly from the .mp4.
The analyst clicks the thumbnail at (or just after) the moment the
barrier opens; the picker then refines with a second tighter grid for
sub-second precision.
"""Interactive picker for barrier-opening time.
Two-stage flow per video:
1. Coarse grid: 60 thumbs spanning the 5-min search window (5 s spacing).
Click pick that 5 s slot.
2. Fine grid: 60 thumbs spanning ±6 s of the coarse pick (0.2 s spacing).
Click final answer with 0.2 s precision.
1. Coarse: 10×6 thumbnail grid spanning 10 min (~10 s spacing) lets
you click the rough moment where the barrier opens.
2. Fine: launches mpv at the coarse pick, paused with on-screen
fractional time. You scrub to the exact frame; on close, type the
time you saw on the OSD into the terminal prompt. Default is the
coarse pick.
After the time is set, the picker also prompts for non-opening ROIs
(comma-separated list, e.g. "1,2,3"). Useful for the partial-opening
videos where only some sub-arenas open. Saved to the `bad_rois` column.
Output: data/metadata/barrier_opening.csv with columns
machine_name, session_date, session_time, opening_s, trim_first_s, notes
machine_name, session_date, session_time, opening_s, trim_first_s,
bad_rois, notes
Window keys:
click select thumbnail at that timestamp
Coarse-grid keys:
click pick that timestamp
n skip this video for THIS run
u mark unusable (opening_s = NaN)
b back to coarse grid (after seeing fine grid)
q / ESC save+quit
Usage:
@ -27,13 +27,16 @@ Usage:
python pick_barrier.py --redo
python pick_barrier.py --limit 10
python pick_barrier.py --db /path/to/specific_tracking.db
python pick_barrier.py --no-player # skip the video-player refinement step
"""
from __future__ import annotations
import argparse
import re
import shutil
import sqlite3
import subprocess
import sys
from pathlib import Path
@ -50,7 +53,7 @@ from detect_barrier_opening import (
OUT_CSV = DATA_METADATA / "barrier_opening.csv"
OUT_COLS = ["machine_name", "session_date", "session_time",
"opening_s", "trim_first_s", "notes"]
"opening_s", "trim_first_s", "bad_rois", "notes"]
DB_NAME_RE = re.compile(
r"^(\d{4}-\d{2}-\d{2})_(\d{2}-\d{2}-\d{2})_([0-9a-f]{32})__"
@ -213,6 +216,96 @@ def show_thumbnail_grid(
return state["time"], state["action"] or "skip"
def parse_time_input(s: str) -> float | None:
"""Accept seconds ('290'), m:ss ('4:50'), or m:ss.ss ('4:50.40')."""
s = s.strip()
if not s:
return None
try:
if ":" in s:
parts = s.split(":")
if len(parts) == 2:
return float(parts[0]) * 60 + float(parts[1])
if len(parts) == 3:
return float(parts[0]) * 3600 + float(parts[1]) * 60 + float(parts[2])
return float(s)
except ValueError:
return None
def play_video(video_path: Path, start_t: float) -> str | None:
"""Launch a video player at start_t-30s. Returns player name, or None.
Tries mpv (best UX for this), then vlc, then ffplay. The user scrubs
to the exact frame, reads the timestamp off the player's OSD/seekbar,
closes the player, and types the time at the terminal prompt.
"""
seek = max(0.0, start_t - 30.0)
if shutil.which("mpv"):
cmd = ["mpv", "--no-resume-playback", "--osd-level=3", "--osd-fractions",
"--pause", f"--start={seek:.1f}", str(video_path)]
name = "mpv"
elif shutil.which("vlc"):
cmd = ["vlc", "--no-video-title-show", f"--start-time={seek:.1f}",
"--play-and-pause", str(video_path)]
name = "vlc"
elif shutil.which("ffplay"):
cmd = ["ffplay", "-hide_banner", "-loglevel", "error",
"-ss", f"{seek:.1f}", str(video_path)]
name = "ffplay"
else:
print(" ! no video player found (tried mpv, vlc, ffplay)")
return None
print(f" launching {name} at {seek:.1f}s — pause on the opening frame, "
"read the time off the player, then close it.")
try:
subprocess.run(cmd, check=False,
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
except OSError as e:
print(f" {name} launch failed: {e}")
return None
return name
def prompt_opening_time(default_s: float) -> float | str:
"""Prompt for the exact opening time. Returns float, 'skip', 'unusable', 'quit'."""
while True:
s = input(
f" opening time [enter={default_s:.1f}s · n=skip · u=unusable · q=quit]: "
).strip()
if not s:
return default_s
if s.lower() in ("n", "skip"):
return "skip"
if s.lower() in ("u", "unusable"):
return "unusable"
if s.lower() in ("q", "quit"):
return "quit"
t = parse_time_input(s)
if t is None:
print(" ? enter seconds, m:ss, or m:ss.ss")
continue
return t
def prompt_bad_rois() -> str:
"""Prompt for non-opening ROIs. Returns comma-separated string (may be empty)."""
while True:
s = input(" non-opening ROIs (e.g. '1,2,3') [enter=none]: ").strip()
if not s:
return ""
# Sanity-check: comma-separated integers in 1..6
try:
rois = sorted({int(x.strip()) for x in s.split(",") if x.strip()})
except ValueError:
print(" ? enter integers separated by commas")
continue
if not all(1 <= r <= 6 for r in rois):
print(" ? ROI numbers must be 1..6")
continue
return ",".join(str(r) for r in rois)
def pick_for_video(
video_path: Path,
db_path: Path | None,
@ -220,60 +313,58 @@ def pick_for_video(
session_date: str,
session_time: str,
coarse_span_s: float = DEFAULT_COARSE_SPAN_S,
use_player: bool = True,
) -> dict | str | None:
"""Run the two-stage thumbnail picker. Return dict, 'skip', or 'quit'."""
"""Run the picker. Return result dict, 'skip', or 'quit'."""
auto_t = auto_suggest(db_path) if db_path else None
print(f" auto-suggest: {f'{auto_t:.1f}s' if auto_t else '(none)'}")
# Stage 1: coarse grid centred on auto-suggest, or middle of span.
# Stage 1: coarse thumbnail grid for rough localisation.
coarse_center = auto_t if auto_t is not None else coarse_span_s / 2
title_coarse = (f"COARSE {machine_name} {session_date} {session_time} "
f"· spanning {coarse_span_s/60:.0f} min")
while True:
coarse_t, action = show_thumbnail_grid(
video_path, coarse_center, coarse_span_s, title_coarse
)
if action == "skip":
return "skip"
if action == "unusable":
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": np.nan,
"trim_first_s": 0, "notes": "unusable",
}
if action == "quit":
return "quit"
if action == "back":
continue # already at the top stage; redraw
if action == "pick" and coarse_t is not None:
break
f"· spanning {coarse_span_s/60:.0f} min "
f"· click ≈ where the barrier opens")
coarse_t, action = show_thumbnail_grid(
video_path, coarse_center, coarse_span_s, title_coarse
)
if action == "skip":
return "skip"
if action == "unusable":
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": np.nan,
"trim_first_s": 0, "bad_rois": "", "notes": "unusable",
}
if action == "quit":
return "quit"
if action != "pick" or coarse_t is None:
return "skip"
# Stage 2: fine grid around the coarse pick.
title_fine = (f"FINE {machine_name} {session_date} {session_time} "
f"· ±{FINE_SPAN_S/2:.0f} s around {coarse_t:.1f} s")
while True:
fine_t, action = show_thumbnail_grid(
video_path, coarse_t, FINE_SPAN_S, title_fine
)
if action == "back":
return pick_for_video(video_path, db_path, machine_name,
session_date, session_time)
if action == "skip":
return "skip"
if action == "unusable":
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": np.nan,
"trim_first_s": 0, "notes": "unusable",
}
if action == "quit":
return "quit"
if action == "pick" and fine_t is not None:
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": round(fine_t, 1),
"trim_first_s": 0, "notes": "",
}
# Stage 2: hand off to a video player for frame-accurate refinement.
if use_player:
played = play_video(video_path, coarse_t)
if played is None:
print(" ! using coarse pick as the answer.")
fine_t = prompt_opening_time(default_s=coarse_t)
if fine_t == "skip":
return "skip"
if fine_t == "unusable":
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": np.nan,
"trim_first_s": 0, "bad_rois": "", "notes": "unusable",
}
if fine_t == "quit":
return "quit"
bad_rois = prompt_bad_rois()
return {
"machine_name": machine_name, "session_date": session_date,
"session_time": session_time, "opening_s": round(float(fine_t), 1),
"trim_first_s": 0, "bad_rois": bad_rois, "notes": "",
}
def lookup_video_path(machine_name: str, session_date: str,
@ -299,6 +390,8 @@ def main() -> None:
help="annotate this specific tracking DB only")
parser.add_argument("--coarse-span", type=float, default=DEFAULT_COARSE_SPAN_S,
help=f"coarse-grid time span in seconds (default {DEFAULT_COARSE_SPAN_S:.0f})")
parser.add_argument("--no-player", action="store_true",
help="skip the video-player refinement step (use the coarse pick directly)")
args = parser.parse_args()
OUT_CSV.parent.mkdir(parents=True, exist_ok=True)
@ -306,6 +399,10 @@ def main() -> None:
out = pd.read_csv(OUT_CSV)
else:
out = pd.DataFrame(columns=OUT_COLS)
# Reason: backfill bad_rois column for older CSVs without it.
for col in OUT_COLS:
if col not in out.columns:
out[col] = ""
done = set(zip(out["machine_name"], out["session_date"], out["session_time"]))
if not INVENTORY_CSV.exists():
@ -363,7 +460,8 @@ def main() -> None:
print(f"\n{prefix}")
result = pick_for_video(video, db, machine_name, session_date, session_time,
coarse_span_s=args.coarse_span)
coarse_span_s=args.coarse_span,
use_player=not args.no_player)
if result is None or result == "skip":
skipped += 1