sungguk's picture
Release visual answerability benchmark v1.0.0
e1ced61 verified
Raw
History Blame Contribute Delete
2.02 kB
"""Direct PlotQA observation comparisons, without digest validation.
These are the production drawing primitives. They are not an independent
renderer; independence applies to the two semantic executors only.
"""
from __future__ import annotations
import copy
import io
from pathlib import Path
_FONT_READY = False
def pixels(blob):
import numpy as np
from PIL import Image
return np.asarray(Image.open(io.BytesIO(blob)).convert("RGBA"))
def render_direct(world, manifest):
global _FONT_READY
import matplotlib
matplotlib.use("Agg")
from matplotlib import font_manager
from explicit_learning.renderers import plot
if not _FONT_READY:
root = Path(matplotlib.get_data_path()) / "fonts" / "ttf"
for name in ("DejaVuSans.ttf", "DejaVuSerif.ttf"):
font_manager.fontManager.addfont(root / name)
_FONT_READY = True
family = plot.PLOT_FAMILIES[manifest["renderer_id"]]
series = plot._parse_series(world)
domain = plot._build_domain(world, series)
canvas = plot._Canvas(manifest["width"], manifest["height"], family)
visible = plot._draw_plot(canvas, world, series, domain, family)
image, owners, counts = canvas.finish()
glyphs = [
{"node_id": g.node_id, "text": g.text, "bbox": [g.x0, g.y0, g.x1, g.y1]}
for g in sorted(canvas.glyphs, key=lambda x: (x.node_id, x.text, x.x0, x.y0))
]
return {
"image": image,
"owners": owners,
"glyphs": glyphs,
"node_table": ["__background__", *canvas.owner_ids],
"coverage": {n: counts.get(n, 0) for n in sorted(visible)},
"domain": domain.manifest_record(),
}
def visible_projection(world):
world = copy.deepcopy(world)
hidden = {n for n, flag in world.get("_node_visibility", {}).items() if flag == "hidden"}
for series in world.get("series", []):
for point in series.get("points", []):
if point["id"] in hidden:
point["y"] = None
return world