File size: 3,812 Bytes
a9d655b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 | """HOWC hierarchical perception - HuggingFace Space (Gradio).
Upload a road image; a detector proposes boxes and each is classified by
hierarchical taxonomic abstraction: the most specific safe level, or an explicit
UNKNOWN, never a confident wrong leaf. Models download lazily on first run.
To deploy: this file plus requirements.txt and README.md, together with the
`hpercept/` package and `taxonomy.yaml` at the Space root (copy them in, or
`pip install` the package). See README.md.
"""
from __future__ import annotations
import numpy as np
import gradio as gr
from hpercept.abstraction import AbstractionConfig
from hpercept.pipeline import get_pipeline
from hpercept.viz import annotate, taxonomy_html
PIPE = get_pipeline()
TAX = PIPE.taxonomy
def analyze(image: np.ndarray, det_conf: float, commit_mass: float, floor: bool):
if image is None:
return None, "Upload an image first.", [], taxonomy_html(TAX, None), gr.update(choices=[])
cfg = AbstractionConfig(commit_mass=float(commit_mass), enforce_floor=bool(floor))
scene = PIPE.run(image, mode="clip", det_conf=float(det_conf), cfg=cfg)
annotated = annotate(image, scene)
rows = [[i, p.classification.label, p.classification.outcome.value,
f"{p.classification.confidence:.2f}", p.box.coco_name]
for i, p in enumerate(scene.predictions)]
c = scene.counts()
summary = (f"**{len(scene.predictions)} detections** - "
f"🟢 {c['identified']} identified · 🟠 {c['abstracted']} abstracted · "
f"🔴 {c['unknown']} unknown · ⬜ {c['rejected']} rejected")
choices = [f"{i}: {p.classification.label}" for i, p in enumerate(scene.predictions)]
first = scene.predictions[0] if scene.predictions else None
analyze.scene = scene
return annotated, summary, rows, taxonomy_html(TAX, first), gr.update(
choices=choices, value=(choices[0] if choices else None))
def highlight(sel: str):
scene = getattr(analyze, "scene", None)
if not scene or not sel:
return taxonomy_html(TAX, None)
i = min(int(sel.split(":")[0]), len(scene.predictions) - 1)
return taxonomy_html(TAX, scene.predictions[i])
with gr.Blocks(title="HOWC Hierarchical Perception") as demo:
gr.Markdown(
"# HOWC : hierarchical perception for novel road objects\n"
"Objects that fit no flat class are **abstracted up a taxonomy** to the "
"most specific *safe* level (bounded by a 🛡 floor), or flagged as an "
"explicit **UNKNOWN OBSTACLE** - never dropped, never a confident wrong "
"label. Paper: doi:10.5281/zenodo.21593472")
with gr.Row():
with gr.Column():
img = gr.Image(label="Road image", type="numpy")
det = gr.Slider(0.05, 0.9, value=0.25, step=0.05, label="Detection confidence")
cm = gr.Slider(0.3, 0.9, value=0.40, step=0.05,
label="Commit mass (↑ = abstracts sooner)")
fl = gr.Checkbox(value=True, label="Safety floor 🛡 (anti-paranoia)")
btn = gr.Button("Analyze", variant="primary")
with gr.Column():
out = gr.Image(label="Annotated", type="numpy")
summ = gr.Markdown()
tbl = gr.Dataframe(headers=["#", "label", "outcome", "conf", "yolo"],
label="Detections")
pick = gr.Dropdown(label="Highlight in taxonomy", choices=[])
tree = gr.HTML(taxonomy_html(TAX, None))
btn.click(analyze, [img, det, cm, fl], [out, summ, tbl, tree, pick])
pick.change(highlight, pick, tree)
if __name__ == "__main__":
import sys
# `python app.py --share` -> a temporary public URL (~72h), for launch-day
# demand tests without hosting a Space.
demo.launch(share="--share" in sys.argv)
|