Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from PIL import Image, ImageDraw, ImageFont
|
| 3 |
+
from transformers import GroundingDinoProcessor
|
| 4 |
+
from modeling_grounding_dino import GroundingDinoForObjectDetection
|
| 5 |
+
|
| 6 |
+
from PIL import Image, ImageDraw, ImageFont
|
| 7 |
+
from itertools import cycle
|
| 8 |
+
|
| 9 |
+
import gradio as gr
|
| 10 |
+
|
| 11 |
+
import spaces
|
| 12 |
+
|
| 13 |
+
# Load model and processor
|
| 14 |
+
model_id = "fushh7/llmdet_swin_large_hf"
|
| 15 |
+
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
| 16 |
+
|
| 17 |
+
print(f"[INFO] Using device: {DEVICE}")
|
| 18 |
+
print(f"[INFO] Loading model from {model_id}...")
|
| 19 |
+
|
| 20 |
+
processor = GroundingDinoProcessor.from_pretrained(model_id)
|
| 21 |
+
model = GroundingDinoForObjectDetection.from_pretrained(model_id).to(DEVICE)
|
| 22 |
+
model.eval();
|
| 23 |
+
|
| 24 |
+
print("[INFO] Model loaded successfully.")
|
| 25 |
+
|
| 26 |
+
# Pre-defined palette (extend or tweak as you like)
|
| 27 |
+
BOX_COLORS = [
|
| 28 |
+
"deepskyblue", "red", "lime", "dodgerblue",
|
| 29 |
+
"cyan", "magenta", "yellow",
|
| 30 |
+
"orange", "chartreuse"
|
| 31 |
+
]
|
| 32 |
+
|
| 33 |
+
def draw_boxes(image, boxes, labels, scores, colors=BOX_COLORS, font_path="arial.ttf", font_size=16):
|
| 34 |
+
"""
|
| 35 |
+
Draw bounding boxes and labels on a PIL Image.
|
| 36 |
+
|
| 37 |
+
:param image: PIL Image object
|
| 38 |
+
:param boxes: Iterable of [x_min, y_min, x_max, y_max]
|
| 39 |
+
:param labels: Iterable of label strings
|
| 40 |
+
:param scores: Iterable of scalar confidences (0-1)
|
| 41 |
+
:param colors: List/tuple of colour names or RGB tuples
|
| 42 |
+
:param font_path: Path to a TTF font for labels
|
| 43 |
+
:param font_size: Int size of font to use, default 16
|
| 44 |
+
:return: PIL Image with drawn boxes
|
| 45 |
+
"""
|
| 46 |
+
# Ensure we can iterate colours indefinitely
|
| 47 |
+
colour_cycle = cycle(colors)
|
| 48 |
+
draw = ImageDraw.Draw(image)
|
| 49 |
+
|
| 50 |
+
# Pick a font (fallback to default if missing)
|
| 51 |
+
try:
|
| 52 |
+
font = ImageFont.truetype(font_path, size=font_size)
|
| 53 |
+
except IOError:
|
| 54 |
+
font = ImageFont.load_default(size=font_size)
|
| 55 |
+
|
| 56 |
+
# Assign a consistent colour per label (optional)
|
| 57 |
+
label_to_colour = {}
|
| 58 |
+
|
| 59 |
+
for box, label, score in zip(boxes, labels, scores):
|
| 60 |
+
# Reuse colour if label seen before, else take next from cycle
|
| 61 |
+
colour = label_to_colour.setdefault(label, next(colour_cycle))
|
| 62 |
+
|
| 63 |
+
x_min, y_min, x_max, y_max = map(int, box)
|
| 64 |
+
|
| 65 |
+
# Draw rectangle
|
| 66 |
+
draw.rectangle([x_min, y_min, x_max, y_max], outline=colour, width=2)
|
| 67 |
+
|
| 68 |
+
# Compose text
|
| 69 |
+
text = f"{label} ({score:.3f})"
|
| 70 |
+
text_size = draw.textbbox((0, 0), text, font=font)[2:]
|
| 71 |
+
|
| 72 |
+
# Draw text background for legibility
|
| 73 |
+
bg_coords = [x_min, y_min - text_size[1] - 4,
|
| 74 |
+
x_min + text_size[0] + 4, y_min]
|
| 75 |
+
draw.rectangle(bg_coords, fill=colour)
|
| 76 |
+
|
| 77 |
+
# Draw text
|
| 78 |
+
draw.text((x_min + 2, y_min - text_size[1] - 2),
|
| 79 |
+
text, fill="black", font=font)
|
| 80 |
+
|
| 81 |
+
return image
|
| 82 |
+
|
| 83 |
+
def resize_image_max_dimension(image, max_size=1024):
|
| 84 |
+
"""
|
| 85 |
+
Resize an image so that the longest side is at most max_size pixels,
|
| 86 |
+
while maintaining the aspect ratio.
|
| 87 |
+
|
| 88 |
+
:param image: PIL Image object
|
| 89 |
+
:param max_size: Maximum dimension in pixels (default: 1024)
|
| 90 |
+
:return: PIL Image object (resized)
|
| 91 |
+
"""
|
| 92 |
+
width, height = image.size
|
| 93 |
+
|
| 94 |
+
# Check if resizing is needed
|
| 95 |
+
if max(width, height) <= max_size:
|
| 96 |
+
return image
|
| 97 |
+
|
| 98 |
+
# Calculate new dimensions maintaining aspect ratio
|
| 99 |
+
ratio = max_size / max(width, height)
|
| 100 |
+
new_width = int(width * ratio)
|
| 101 |
+
new_height = int(height * ratio)
|
| 102 |
+
|
| 103 |
+
# Resize the image using high-quality resampling
|
| 104 |
+
return image.resize((new_width, new_height), Image.Resampling.LANCZOS)
|
| 105 |
+
|
| 106 |
+
@spaces.GPU(duration=120)
|
| 107 |
+
def detect_and_draw(
|
| 108 |
+
img: Image.Image,
|
| 109 |
+
text_query: str,
|
| 110 |
+
box_threshold: float = 0.4,
|
| 111 |
+
text_threshold: float = 0.3
|
| 112 |
+
) -> Image.Image:
|
| 113 |
+
"""
|
| 114 |
+
Detect objects described in `text_query`, draw boxes, return the image.
|
| 115 |
+
Note: `text_query` must be lowercase and each concept ends with a dot
|
| 116 |
+
(e.g. 'a cat. a remote control.')
|
| 117 |
+
"""
|
| 118 |
+
|
| 119 |
+
# Make sure text is lowered
|
| 120 |
+
text_query = text_query.lower()
|
| 121 |
+
|
| 122 |
+
# If the image size is too large, we make it smaller
|
| 123 |
+
img = resize_image_max_dimension(img, max_size=1024)
|
| 124 |
+
|
| 125 |
+
# Preprocess the image
|
| 126 |
+
inputs = processor(images=img, text=text_query, return_tensors="pt").to(DEVICE)
|
| 127 |
+
|
| 128 |
+
with torch.no_grad():
|
| 129 |
+
outputs = model(**inputs)
|
| 130 |
+
|
| 131 |
+
results = processor.post_process_grounded_object_detection(
|
| 132 |
+
outputs,
|
| 133 |
+
inputs.input_ids,
|
| 134 |
+
box_threshold=box_threshold,
|
| 135 |
+
text_threshold=text_threshold,
|
| 136 |
+
target_sizes=[img.size[::-1]]
|
| 137 |
+
)[0]
|
| 138 |
+
|
| 139 |
+
img_out = img.copy()
|
| 140 |
+
img_out = draw_boxes(
|
| 141 |
+
img_out,
|
| 142 |
+
boxes = results["boxes"].cpu().numpy(),
|
| 143 |
+
labels = results.get("text_labels", results.get("labels", [])),
|
| 144 |
+
scores = results["scores"]
|
| 145 |
+
)
|
| 146 |
+
return img_out
|
| 147 |
+
|
| 148 |
+
# Create Gradio demo
|
| 149 |
+
demo = gr.Interface(
|
| 150 |
+
fn = detect_and_draw,
|
| 151 |
+
inputs = [
|
| 152 |
+
gr.Image(type="pil", label="Image"),
|
| 153 |
+
gr.Textbox(value="",
|
| 154 |
+
label="Text Query (lowercase, end each with '.', for example 'a bird. a tree.')"),
|
| 155 |
+
gr.Slider(0.0, 1.0, 0.4, 0.05, label="Box Threshold"),
|
| 156 |
+
gr.Slider(0.0, 1.0, 0.3, 0.05, label="Text Threshold")
|
| 157 |
+
],
|
| 158 |
+
outputs = gr.Image(type="pil", label="Detections"),
|
| 159 |
+
title = "LLMDet Demo: Open-Vocabulary Grounded Object Detection",
|
| 160 |
+
description = """Upload an image, enter text queries, and adjust thresholds to see detections.
|
| 161 |
+
|
| 162 |
+
Adapted from LLMDet GitHub repo [Hugging Face demo](https://github.com/iSEE-Laboratory/LLMDet/tree/main/hf_model).
|
| 163 |
+
|
| 164 |
+
See original:
|
| 165 |
+
* [LLMDet GitHub](https://github.com/iSEE-Laboratory/LLMDet/tree/main?tab=readme-ov-file)
|
| 166 |
+
* [LLMDet Paper](https://arxiv.org/abs/2501.18954) - LLMDet: Learning Strong Open-Vocabulary Object Detectors under the Supervision of Large Language Models
|
| 167 |
+
* [LLMDet model checkpoint](https://huggingface.co/fushh7/llmdet_swin_large_hf)
|
| 168 |
+
"""
|
| 169 |
+
)
|
| 170 |
+
|
| 171 |
+
demo.launch()
|