activevision's picture
Initial release v0.4.0 — ActiveVision benchmark (85 instances, 17 tasks)
f69e256 verified
raw
history blame
19.6 kB
"""Generate 'Spot the Contour Diff' visual benchmark dataset.
Two panels are shown side by side. Each panel contains the same set of closed
contours (Fourier-descriptor blobs) at the same positions, with each contour
pair tagged by an uppercase letter label that appears at the same position in
both panels. Some contours in the right panel have a different shape from the
corresponding contour in the left panel. The task is to list the letter labels
of contour pairs whose shape differs between the two panels.
Blob generation is reused from the sibling ``contour_silhouette_count`` task.
"""
from __future__ import annotations
import argparse
import json
import math
import random
import sys
from pathlib import Path
from typing import Any, Dict, List, Tuple
import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import numpy as np
from tqdm import tqdm
# Reuse blob generation and stroke helpers from contour_silhouette_count.
_SIBLING = Path(__file__).resolve().parent.parent / "contour_silhouette_count"
sys.path.insert(0, str(_SIBLING))
from creation import ( # type: ignore # noqa: E402
STROKE_WIDTH,
_draw_blob_outline,
blob_bounding_radius,
blob_hausdorff,
fourier_blob,
)
# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------
NUM_ITEMS = 15
MIN_DIFFS = 4
MAX_DIFFS = 7
PANEL_W = 900
PANEL_H = 900
PADDING = 18 # visible gap between blob outlines
EDGE_PADDING = 18 # gap between blob bounding circle and panel edge
HAUSDORFF_THRESH = 5.0 # changed blob must differ from original by at least this
HAUSDORFF_MAX = 20.0 # …but not by more than this (keep diffs subtle)
BG_COLOR = "#0a1020"
BORDER_COLOR = "#7aa6ff"
BORDER_WIDTH = 2.0
HIGHLIGHT_COLOR = "#dc1e1e"
LABEL_BADGE_COLOR = "#f5d76e"
LABEL_TEXT_COLOR = "#1a1a1a"
LABEL_BADGE_PAD = 8.0 # gap between bounding circle and badge centre offset
MARGIN_PX = 60
GAP_PX = 100
LABEL_HEIGHT = 44
QUESTION = (
"Two panels are shown side by side. Each panel contains the same set of "
"closed contours at the same positions. Some contours in the right "
"panel have a different shape from the corresponding contour in the "
"left panel. Count the number of contour pairs whose shape differs "
"between the two panels and report the integer count. "
"Provide your final answer enclosed in <answer>...</answer> tags."
)
def _index_to_label(i: int) -> str:
"""Map 0->A, 1->B, ..., 25->Z, 26->AA, 27->AB, ..."""
letters = []
n = i
while True:
letters.append(chr(ord("A") + (n % 26)))
n = n // 26 - 1
if n < 0:
break
return "".join(reversed(letters))
# ---------------------------------------------------------------------------
# Blob generation / swapping
# ---------------------------------------------------------------------------
def _generate_blob(rng: random.Random) -> np.ndarray:
r0 = rng.uniform(38.0, 52.0)
return fourier_blob(rng, r0=r0, K_range=(4, 6), amp_scale=0.22)
def _polygon_area(poly: np.ndarray) -> float:
x = poly[:, 0]
y = poly[:, 1]
return 0.5 * abs(float(np.dot(x, np.roll(y, -1)) - np.dot(y, np.roll(x, -1))))
# Minimum fractional difference in enclosed area between the original and the
# swapped contour. Together with HAUSDORFF_THRESH this guarantees that the
# swap differs both in OUTLINE shape and in ENCLOSED REGION.
MIN_AREA_RATIO_DIFF = 0.18
def _sample_different_blob(
rng: random.Random,
old: np.ndarray,
old_radius: float,
max_tries: int = 800,
) -> np.ndarray:
"""Sample a new blob with Hausdorff distance >= HAUSDORFF_THRESH from
``old``, area differing by at least MIN_AREA_RATIO_DIFF, and bounding
radius no larger than ``old_radius``."""
old_area = _polygon_area(old)
for _ in range(max_tries):
cand = _generate_blob(rng)
if blob_bounding_radius(cand) > old_radius:
continue
d = blob_hausdorff(cand, old)
if not (HAUSDORFF_THRESH <= d <= HAUSDORFF_MAX):
continue
cand_area = _polygon_area(cand)
if old_area > 0 and abs(cand_area - old_area) / old_area < MIN_AREA_RATIO_DIFF:
continue
return cand
raise RuntimeError("Could not sample a sufficiently different blob.")
# ---------------------------------------------------------------------------
# Layout
# ---------------------------------------------------------------------------
def _place_items(
rng: random.Random,
radii: List[float],
max_tries_per_item: int = 5000,
) -> List[Tuple[float, float]]:
order = sorted(range(len(radii)), key=lambda i: -radii[i])
centers: List[Tuple[float, float] | None] = [None] * len(radii)
for i in order:
r = radii[i]
placed = False
for _ in range(max_tries_per_item):
x = rng.uniform(r + EDGE_PADDING, PANEL_W - r - EDGE_PADDING)
y = rng.uniform(r + EDGE_PADDING, PANEL_H - r - EDGE_PADDING)
ok = True
for j, c in enumerate(centers):
if c is None or j == i:
continue
min_dist = radii[i] + radii[j] + PADDING
dx = x - c[0]
dy = y - c[1]
if dx * dx + dy * dy < min_dist * min_dist:
ok = False
break
if ok:
centers[i] = (x, y)
placed = True
break
if not placed:
raise RuntimeError(f"Could not place blob {i} (r={r:.1f}) without overlap.")
return [c for c in centers] # type: ignore[return-value]
# ---------------------------------------------------------------------------
# Sample construction
# ---------------------------------------------------------------------------
def build_sample(
rng: random.Random,
num_items: int,
num_diffs: int,
) -> Tuple[List[Dict[str, Any]], List[Dict[str, Any]]]:
left_blobs = [_generate_blob(rng) for _ in range(num_items)]
radii = [blob_bounding_radius(b) for b in left_blobs]
right_blobs: List[np.ndarray] = [b.copy() for b in left_blobs]
changed_idx = rng.sample(range(num_items), num_diffs)
for i in changed_idx:
right_blobs[i] = _sample_different_blob(rng, left_blobs[i], radii[i])
centers = _place_items(rng, radii)
changed_set = set(changed_idx)
items: List[Dict[str, Any]] = []
for i in range(num_items):
x, y = centers[i]
items.append({
"index": i,
"label": _index_to_label(i),
"x": x,
"y": y,
"bounding_radius": radii[i],
"left_pts": left_blobs[i],
"right_pts": right_blobs[i],
"changed": i in changed_set,
})
diffs: List[Dict[str, Any]] = [
{
"index": i,
"label": items[i]["label"],
"x": items[i]["x"],
"y": items[i]["y"],
"bounding_radius": items[i]["bounding_radius"],
}
for i in sorted(changed_idx)
]
return items, diffs
# ---------------------------------------------------------------------------
# Rendering
# ---------------------------------------------------------------------------
def _panel_origins() -> Tuple[Tuple[float, float], Tuple[float, float]]:
oy = MARGIN_PX + LABEL_HEIGHT
ox_left = MARGIN_PX
ox_right = MARGIN_PX + PANEL_W + GAP_PX
return (ox_left, oy), (ox_right, oy)
def _canvas_size() -> Tuple[int, int]:
w = MARGIN_PX + PANEL_W + GAP_PX + PANEL_W + MARGIN_PX
h = MARGIN_PX + LABEL_HEIGHT + PANEL_H + MARGIN_PX
return w, h
def _badge_offset(
x: float,
y: float,
radius: float,
badge_r: float,
) -> Tuple[float, float]:
"""Legacy compass-based fallback (unused — kept for backward compatibility)."""
margin = radius + badge_r + LABEL_BADGE_PAD
candidates = [
(1.0, -1.0), (1.0, 1.0), (-1.0, -1.0), (-1.0, 1.0),
(1.0, 0.0), (0.0, -1.0), (-1.0, 0.0), (0.0, 1.0),
]
inset = badge_r + 2.0
for sx, sy in candidates:
norm = math.hypot(sx, sy) or 1.0
dx = sx / norm * margin
dy = sy / norm * margin
bx = x + dx
by = y + dy
if (inset <= bx <= PANEL_W - inset
and inset <= by <= PANEL_H - inset):
return dx, dy
return margin / math.sqrt(2), -margin / math.sqrt(2)
def _plan_label_positions(
items: List[Dict[str, Any]],
other_clearance: float = 14.0,
label_clearance: float = 24.0,
n_angles: int = 16,
radial_steps: Tuple[float, ...] = (0.0, 6.0, 14.0, 24.0),
) -> List[Tuple[float, float]]:
"""Choose label-badge centres (in LEFT-panel local coords) using a
multi-candidate, best-score search.
Returns one (px, py) per item. Identical coordinates are reused for the
right panel (just translated by that panel's origin) since both panels
have the same width/height and contour positions.
"""
n = len(items)
# Pre-compute badge radii and translated point sets (panel-local).
badge_rs: List[float] = []
other_pts_list: List[np.ndarray] = []
for it in items:
label = it["label"]
badge_rs.append(22.0 + 6.0 * (len(label) - 1))
# Translate the LEFT contour's points into panel-local coords.
other_pts_list.append(it["left_pts"] + np.array([it["x"], it["y"]]))
placements: List[Tuple[float, float]] = []
placed_centres: List[Tuple[float, float]] = []
for i, it in enumerate(items):
x, y = it["x"], it["y"]
radius = it["bounding_radius"]
badge_r = badge_rs[i]
inset = badge_r + 2.0
base_margin = radius + badge_r + LABEL_BADGE_PAD
# Other contours' points (everything except self).
if n > 1:
other_pts = np.concatenate(
[other_pts_list[j] for j in range(n) if j != i], axis=0
)
else:
other_pts = np.zeros((0, 2))
def evaluate(oc: float, lc: float):
best = None
best_score = -1e18
for a_idx in range(n_angles):
ang = 2.0 * math.pi * a_idx / n_angles
cos_a = math.cos(ang)
sin_a = math.sin(ang)
for off in radial_steps:
margin = base_margin + off
px = x + cos_a * margin
py = y + sin_a * margin
# Panel-bounds gate.
if not (inset <= px <= PANEL_W - inset
and inset <= py <= PANEL_H - inset):
continue
# Other-contour clearance.
if other_pts.shape[0] > 0:
dx = other_pts[:, 0] - px
dy = other_pts[:, 1] - py
other_d = float(np.sqrt(np.min(dx * dx + dy * dy)))
else:
other_d = 1e6
if other_d < oc:
continue
# Label-clearance gate.
if placed_centres:
label_d = min(
math.hypot(px - lx, py - ly)
for lx, ly in placed_centres
)
else:
label_d = 1e6
if label_d < lc:
continue
score = other_d + 0.5 * label_d - 0.4 * off
if score > best_score:
best_score = score
best = (px, py)
return best
chosen = evaluate(other_clearance, label_clearance)
# Soft fallback: progressively relax other_clearance.
relax = other_clearance
while chosen is None and relax > 1.0:
relax *= 0.6
chosen = evaluate(relax, max(4.0, label_clearance * 0.6))
if chosen is None:
# Last-resort: legacy compass.
dx, dy = _badge_offset(x, y, radius, badge_r)
chosen = (x + dx, y + dy)
placements.append(chosen)
placed_centres.append(chosen)
return placements
def _draw_panel(
ax: plt.Axes,
items: List[Dict[str, Any]],
side: str,
ox: float,
oy: float,
label_positions: List[Tuple[float, float]],
) -> None:
key = "left_pts" if side == "left" else "right_pts"
for it in items:
translated = it[key] + np.array([ox + it["x"], oy + it["y"]])
_draw_blob_outline(ax, translated, STROKE_WIDTH, color="white", zorder=3)
# Letter-label badges removed: the task is now a count, not a label-list,
# so explicit per-contour anchors are no longer needed (and would let a
# model match in text space rather than visually).
def _render(
out_path: Path,
items: List[Dict[str, Any]],
diffs: List[Dict[str, Any]] | None = None,
) -> None:
w, h = _canvas_size()
dpi = 100
fig, ax = plt.subplots(1, 1, figsize=(w / dpi, h / dpi), dpi=dpi)
ax.set_xlim(0, w)
ax.set_ylim(h, 0)
ax.set_aspect("equal")
ax.axis("off")
fig.patch.set_facecolor(BG_COLOR)
ax.set_facecolor(BG_COLOR)
(ox_left, oy), (ox_right, _) = _panel_origins()
for ox in (ox_left, ox_right):
border = mpatches.Rectangle(
(ox, oy), PANEL_W, PANEL_H,
facecolor="none", edgecolor=BORDER_COLOR,
linewidth=BORDER_WIDTH, zorder=1,
)
ax.add_patch(border)
label_positions = _plan_label_positions(items)
_draw_panel(ax, items, "left", ox_left, oy, label_positions)
_draw_panel(ax, items, "right", ox_right, oy, label_positions)
ax.text(
ox_left + PANEL_W / 2, MARGIN_PX + LABEL_HEIGHT * 0.5,
"Left", ha="center", va="center",
fontsize=16, fontweight="bold", color="#cfe0ff",
)
ax.text(
ox_right + PANEL_W / 2, MARGIN_PX + LABEL_HEIGHT * 0.5,
"Right", ha="center", va="center",
fontsize=16, fontweight="bold", color="#cfe0ff",
)
if diffs:
for diff in diffs:
hl_r = diff["bounding_radius"] + 12
for ox in (ox_left, ox_right):
cx = ox + diff["x"]
cy = oy + diff["y"]
ring = mpatches.Circle(
(cx, cy), hl_r,
facecolor="none", edgecolor=HIGHLIGHT_COLOR,
linewidth=2.5, zorder=10,
)
ax.add_patch(ring)
fig.savefig(out_path, facecolor=BG_COLOR)
plt.close(fig)
def render_pair(out_path: Path, items: List[Dict[str, Any]]) -> None:
_render(out_path, items)
def render_answer(
out_path: Path,
items: List[Dict[str, Any]],
diffs: List[Dict[str, Any]],
) -> None:
_render(out_path, items, diffs)
# ---------------------------------------------------------------------------
# Annotation
# ---------------------------------------------------------------------------
def _answer_string(diffs: List[Dict[str, Any]]) -> str:
return str(len(diffs))
def build_annotation(
image_name: str,
items: List[Dict[str, Any]],
diffs: List[Dict[str, Any]],
) -> Dict[str, Any]:
return {
"image": image_name,
"num_items": len(items),
"num_differences": len(diffs),
"differences": [
{
"index": d["index"],
"label": d["label"],
"x": d["x"],
"y": d["y"],
"bounding_radius": d["bounding_radius"],
}
for d in diffs
],
"question": QUESTION,
"answer": _answer_string(diffs),
}
# ---------------------------------------------------------------------------
# Dataset generation
# ---------------------------------------------------------------------------
def generate_dataset(
rng: random.Random,
count: int,
output_dir: Path,
num_items: int = NUM_ITEMS,
min_diffs: int = MIN_DIFFS,
max_diffs: int = MAX_DIFFS,
) -> None:
images_dir = output_dir / "images"
answers_dir = output_dir / "answers"
images_dir.mkdir(parents=True, exist_ok=True)
answers_dir.mkdir(parents=True, exist_ok=True)
annotations: List[Dict[str, Any]] = []
data_items: List[Dict[str, Any]] = []
# Force evenly-spaced num_diffs across [min_diffs, max_diffs].
if count > 1:
forced = [int(round(min_diffs + i * (max_diffs - min_diffs) / (count - 1))) for i in range(count)]
else:
forced = [min_diffs]
print(f"forced contour diff counts: {forced}")
for idx in tqdm(range(count), desc="Generating contour diff pairs"):
num_diffs = forced[idx]
for attempt in range(200):
try:
items, diffs = build_sample(rng, num_items, num_diffs)
break
except RuntimeError:
continue
else:
raise RuntimeError(f"Failed to build sample {idx} after many retries")
image_name = f"contour_diff_{idx:05d}.png"
img_path = images_dir / image_name
ans_path = answers_dir / image_name
render_pair(img_path, items)
render_answer(ans_path, items, diffs)
rel_image = f"images/{image_name}"
annotations.append(build_annotation(rel_image, items, diffs))
data_items.append({
"image": rel_image,
"question": QUESTION,
"answer": _answer_string(diffs),
})
with (output_dir / "annotations.jsonl").open("w", encoding="utf-8") as fh:
for rec in annotations:
fh.write(json.dumps(rec) + "\n")
data_json = {
"task": "spot_the_contour_diff",
"category": "visual_attribute_transfer",
"count": len(data_items),
"items": data_items,
}
with (output_dir / "data.json").open("w", encoding="utf-8") as fh:
json.dump(data_json, fh, indent=2)
fh.write("\n")
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def parse_args() -> argparse.Namespace:
p = argparse.ArgumentParser(
description="Generate 'Spot the Contour Diff' visual benchmark dataset."
)
p.add_argument("--output-root", type=Path, default=".")
p.add_argument("--count", type=int, default=20)
p.add_argument("--seed", type=int, default=42)
p.add_argument("--difficulty", type=int, default=5,
help="Integer difficulty >=0; scales diff count and subtlety.")
return p.parse_args()
def main() -> None:
args = parse_args()
rng = random.Random(args.seed)
d = max(0, int(args.difficulty))
N_d = 10 + 2 * d
N_0 = 10
s = math.sqrt(max(1.0, N_d / N_0))
global PANEL_W, PANEL_H
PANEL_W = int(round(PANEL_W * s))
PANEL_H = int(round(PANEL_H * s))
global HAUSDORFF_THRESH, HAUSDORFF_MAX
HAUSDORFF_THRESH = max(18.0, 22.0 - 0.4 * d)
HAUSDORFF_MAX = max(HAUSDORFF_THRESH + 12.0, 50.0 - 1.0 * d)
total_contours = 10 + 2 * d
min_diffs = 5
max_diffs = 5 + 2 * d
generate_dataset(rng, args.count, args.output_root,
num_items=total_contours,
min_diffs=min_diffs, max_diffs=max_diffs)
print(f"Saved {args.count} image pairs to {args.output_root}")
if __name__ == "__main__":
main()