Spaces:
Running on Zero
Running on Zero
Upload folder using huggingface_hub
Browse files- README.md +18 -7
- __pycache__/app.cpython-311.pyc +0 -0
- app.py +139 -0
- requirements.txt +4 -0
README.md
CHANGED
|
@@ -1,13 +1,24 @@
|
|
| 1 |
---
|
| 2 |
-
title:
|
| 3 |
-
emoji:
|
| 4 |
-
colorFrom:
|
| 5 |
-
colorTo:
|
| 6 |
sdk: gradio
|
| 7 |
sdk_version: 6.20.0
|
| 8 |
-
python_version: '3.12'
|
| 9 |
app_file: app.py
|
| 10 |
-
|
|
|
|
|
|
|
| 11 |
---
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
title: CyanBridge Walking Aid
|
| 3 |
+
emoji: "\U0001F9ED"
|
| 4 |
+
colorFrom: blue
|
| 5 |
+
colorTo: indigo
|
| 6 |
sdk: gradio
|
| 7 |
sdk_version: 6.20.0
|
|
|
|
| 8 |
app_file: app.py
|
| 9 |
+
short_description: YOLO and depth guidance for walking aid research
|
| 10 |
+
python_version: "3.12"
|
| 11 |
+
startup_duration_timeout: 30m
|
| 12 |
---
|
| 13 |
|
| 14 |
+
# CyanBridge Walking Aid
|
| 15 |
+
|
| 16 |
+
An experimental accessibility-assistance demo that combines local inference models:
|
| 17 |
+
|
| 18 |
+
- YOLO11n for COCO object detection.
|
| 19 |
+
- YOLO-World for user-provided open-vocabulary object prompts.
|
| 20 |
+
- Depth Anything V2 Small for relative-depth guidance.
|
| 21 |
+
|
| 22 |
+
The output is advisory only. It is not a replacement for a cane, guide dog, mobility training, awareness, or professional advice.
|
| 23 |
+
|
| 24 |
+
The `/api/analyze` endpoint is reserved for the CyanBridge Pro relay. It requires its server-side `X-CyanBridge-Key` header and is never called with a client-side secret.
|
__pycache__/app.cpython-311.pyc
ADDED
|
Binary file (11.9 kB). View file
|
|
|
app.py
ADDED
|
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import base64
|
| 2 |
+
import io
|
| 3 |
+
import os
|
| 4 |
+
from typing import Any
|
| 5 |
+
|
| 6 |
+
import gradio as gr
|
| 7 |
+
import spaces
|
| 8 |
+
import torch
|
| 9 |
+
from fastapi import Header, HTTPException
|
| 10 |
+
from huggingface_hub import hf_hub_download
|
| 11 |
+
from PIL import Image, ImageDraw
|
| 12 |
+
from transformers import AutoImageProcessor, AutoModelForDepthEstimation
|
| 13 |
+
from ultralytics import YOLO
|
| 14 |
+
|
| 15 |
+
YOLO11_PATH = hf_hub_download("Ultralytics/YOLO11", "yolo11n.pt")
|
| 16 |
+
YOLO_WORLD_PATH = hf_hub_download("Bingsu/yolo-world-mirror", "yolov8s-worldv2.pt")
|
| 17 |
+
DEPTH_MODEL_ID = "depth-anything/Depth-Anything-V2-Small-hf"
|
| 18 |
+
|
| 19 |
+
YOLO11 = YOLO(YOLO11_PATH)
|
| 20 |
+
YOLO_WORLD = YOLO(YOLO_WORLD_PATH)
|
| 21 |
+
DEPTH_PROCESSOR = AutoImageProcessor.from_pretrained(DEPTH_MODEL_ID)
|
| 22 |
+
DEPTH_MODEL = AutoModelForDepthEstimation.from_pretrained(DEPTH_MODEL_ID)
|
| 23 |
+
SPACE_KEY = os.environ.get("CYANBRIDGE_SPACE_KEY", "")
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def _device() -> str:
|
| 27 |
+
return "cuda" if torch.cuda.is_available() else "cpu"
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
def _depth_summary(image: Image.Image) -> tuple[Image.Image, dict[str, Any]]:
|
| 31 |
+
device = _device()
|
| 32 |
+
DEPTH_MODEL.to(device)
|
| 33 |
+
inputs = DEPTH_PROCESSOR(images=image, return_tensors="pt").to(device)
|
| 34 |
+
with torch.inference_mode():
|
| 35 |
+
predicted_depth = DEPTH_MODEL(**inputs).predicted_depth
|
| 36 |
+
depth = torch.nn.functional.interpolate(
|
| 37 |
+
predicted_depth.unsqueeze(1),
|
| 38 |
+
size=(image.height, image.width),
|
| 39 |
+
mode="bicubic",
|
| 40 |
+
align_corners=False,
|
| 41 |
+
).squeeze().float().cpu()
|
| 42 |
+
normalized = (depth - depth.min()) / (depth.max() - depth.min() + 1e-6)
|
| 43 |
+
depth_image = Image.fromarray((normalized.numpy() * 255).astype("uint8"))
|
| 44 |
+
lower_center = normalized[int(image.height * 0.6):, int(image.width * 0.33):int(image.width * 0.67)]
|
| 45 |
+
return depth_image, {
|
| 46 |
+
"relative_depth": "Higher values indicate relatively closer surfaces.",
|
| 47 |
+
"center_ground_mean": round(float(lower_center.mean()), 3),
|
| 48 |
+
"center_ground_range": round(float(lower_center.max() - lower_center.min()), 3),
|
| 49 |
+
}
|
| 50 |
+
|
| 51 |
+
|
| 52 |
+
def _detections(image: Image.Image, detector: str, classes: str) -> tuple[Image.Image, list[dict[str, Any]]]:
|
| 53 |
+
model = YOLO_WORLD if detector == "YOLO-World" else YOLO11
|
| 54 |
+
if detector == "YOLO-World":
|
| 55 |
+
prompts = [item.strip() for item in classes.split(",") if item.strip()]
|
| 56 |
+
if prompts:
|
| 57 |
+
model.set_classes(prompts)
|
| 58 |
+
results = model.predict(image, device=_device(), verbose=False, conf=0.35)
|
| 59 |
+
result = results[0]
|
| 60 |
+
annotated = Image.fromarray(result.plot()[..., ::-1])
|
| 61 |
+
names = result.names
|
| 62 |
+
output: list[dict[str, Any]] = []
|
| 63 |
+
for box in result.boxes:
|
| 64 |
+
class_id = int(box.cls.item())
|
| 65 |
+
output.append(
|
| 66 |
+
{
|
| 67 |
+
"label": str(names[class_id]),
|
| 68 |
+
"confidence": round(float(box.conf.item()), 3),
|
| 69 |
+
"box": [round(float(value), 1) for value in box.xyxy[0].tolist()],
|
| 70 |
+
}
|
| 71 |
+
)
|
| 72 |
+
return annotated, output
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
@spaces.GPU(duration=60)
|
| 76 |
+
def analyze_image(image: Image.Image, detector: str, classes: str) -> tuple[Image.Image, Image.Image, dict[str, Any]]:
|
| 77 |
+
"""Detect objects and estimate relative depth for an uploaded walking-scene image."""
|
| 78 |
+
if image is None:
|
| 79 |
+
raise gr.Error("Upload an image first.")
|
| 80 |
+
image = image.convert("RGB")
|
| 81 |
+
detected_image, detected_objects = _detections(image, detector, classes)
|
| 82 |
+
depth_image, depth = _depth_summary(image)
|
| 83 |
+
return detected_image, depth_image, {"detector": detector, "objects": detected_objects, "depth": depth}
|
| 84 |
+
|
| 85 |
+
|
| 86 |
+
def _authorized(key: str | None) -> None:
|
| 87 |
+
if not SPACE_KEY or key != SPACE_KEY:
|
| 88 |
+
raise HTTPException(status_code=401, detail="relay authorization required")
|
| 89 |
+
|
| 90 |
+
|
| 91 |
+
@spaces.GPU(duration=60)
|
| 92 |
+
def analyze_relay_image(image: Image.Image, detector: str, classes: str) -> dict[str, Any]:
|
| 93 |
+
"""Run the model stack for an authorized CyanBridge Pro relay request."""
|
| 94 |
+
_, detected_objects = _detections(image.convert("RGB"), detector, classes)
|
| 95 |
+
_, depth = _depth_summary(image.convert("RGB"))
|
| 96 |
+
return {"detector": detector, "objects": detected_objects, "depth": depth}
|
| 97 |
+
|
| 98 |
+
|
| 99 |
+
with gr.Blocks(title="CyanBridge Walking Aid") as demo:
|
| 100 |
+
gr.Markdown(
|
| 101 |
+
"# CyanBridge Walking Aid\n"
|
| 102 |
+
"Experimental object detection and relative-depth guidance. Do not rely on this demo as your sole mobility aid."
|
| 103 |
+
)
|
| 104 |
+
with gr.Row():
|
| 105 |
+
image_input = gr.Image(type="pil", label="Walking scene")
|
| 106 |
+
with gr.Column():
|
| 107 |
+
detector_input = gr.Radio(["YOLO11n", "YOLO-World"], value="YOLO11n", label="Detector")
|
| 108 |
+
classes_input = gr.Textbox(value="person, bicycle, curb, stairs, pothole", label="YOLO-World classes")
|
| 109 |
+
run_button = gr.Button("Analyze scene", variant="primary")
|
| 110 |
+
with gr.Row():
|
| 111 |
+
detection_output = gr.Image(label="Object detection")
|
| 112 |
+
depth_output = gr.Image(label="Relative depth")
|
| 113 |
+
summary_output = gr.JSON(label="Advisory result")
|
| 114 |
+
run_button.click(
|
| 115 |
+
analyze_image,
|
| 116 |
+
inputs=[image_input, detector_input, classes_input],
|
| 117 |
+
outputs=[detection_output, depth_output, summary_output],
|
| 118 |
+
api_name="analyze",
|
| 119 |
+
)
|
| 120 |
+
|
| 121 |
+
|
| 122 |
+
@demo.app.post("/api/analyze")
|
| 123 |
+
def analyze_relay(payload: dict[str, Any], x_cyanbridge_key: str | None = Header(default=None)) -> dict[str, Any]:
|
| 124 |
+
_authorized(x_cyanbridge_key)
|
| 125 |
+
encoded = str(payload.get("image_base64", ""))
|
| 126 |
+
if not encoded:
|
| 127 |
+
raise HTTPException(status_code=400, detail="image_base64 is required")
|
| 128 |
+
try:
|
| 129 |
+
image = Image.open(io.BytesIO(base64.b64decode(encoded))).convert("RGB")
|
| 130 |
+
except Exception as error:
|
| 131 |
+
raise HTTPException(status_code=400, detail="invalid image payload") from error
|
| 132 |
+
detector = str(payload.get("detector", "YOLO11n"))
|
| 133 |
+
if detector not in {"YOLO11n", "YOLO-World"}:
|
| 134 |
+
raise HTTPException(status_code=400, detail="unsupported detector")
|
| 135 |
+
classes = str(payload.get("classes", ""))
|
| 136 |
+
return analyze_relay_image(image, detector, classes)
|
| 137 |
+
|
| 138 |
+
|
| 139 |
+
demo.launch(mcp_server=True)
|
requirements.txt
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
ultralytics>=8.3.0
|
| 2 |
+
transformers>=4.48.0
|
| 3 |
+
torchvision
|
| 4 |
+
pillow
|