File size: 4,255 Bytes
bbc53fb d294c67 bbc53fb | 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 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 | import os
import numpy as np
import torch
import torch.nn.functional as F
import gradio as gr
from PIL import Image
from transformers import SegformerForSemanticSegmentation, SegformerImageProcessor
# ============================================================
# Config
# ============================================================
MODEL_ID = "Dinusharg/segformer_environment_1"
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# Class colors for your labels
PALETTE = {
0: (0, 0, 0), # background
1: (255, 0, 0), # footpath
2: (0, 255, 0), # grass
3: (128, 128, 128), # road
4: (0, 0, 255), # water
}
# Optional nicer class names for display
CLASS_NAMES = {
0: "background",
1: "footpath",
2: "grass",
3: "road",
4: "water",
}
# ============================================================
# Load model once at startup
# ============================================================
processor = SegformerImageProcessor.from_pretrained(MODEL_ID)
model = SegformerForSemanticSegmentation.from_pretrained(MODEL_ID).to(DEVICE)
model.eval()
def make_color_mask(pred: np.ndarray) -> np.ndarray:
"""
Convert class-id mask [H, W] into RGB color mask [H, W, 3].
"""
color_mask = np.zeros((pred.shape[0], pred.shape[1], 3), dtype=np.uint8)
for class_id, color in PALETTE.items():
color_mask[pred == class_id] = color
return color_mask
def make_overlay(image_np: np.ndarray, color_mask: np.ndarray, alpha: float = 0.5) -> np.ndarray:
"""
Blend original image with color mask.
"""
image_f = image_np.astype(np.float32)
mask_f = color_mask.astype(np.float32)
overlay = (image_f * (1.0 - alpha) + mask_f * alpha).clip(0, 255).astype(np.uint8)
return overlay
def make_class_summary(pred: np.ndarray) -> str:
"""
Summarize predicted pixel counts and percentages.
"""
total_pixels = pred.size
unique, counts = np.unique(pred, return_counts=True)
lines = []
for class_id, count in zip(unique.tolist(), counts.tolist()):
pct = 100.0 * count / total_pixels
class_name = CLASS_NAMES.get(class_id, f"class_{class_id}")
lines.append(f"- {class_id}: {class_name} — {count} pixels ({pct:.2f}%)")
return "\n".join(lines)
def predict(image: Image.Image):
"""
Gradio callback.
Input: PIL image
Outputs:
- segmentation mask
- overlay
- class summary
"""
if image is None:
return None, None, "No image provided."
image = image.convert("RGB")
original_np = np.array(image)
inputs = processor(images=image, return_tensors="pt")
inputs = {k: v.to(DEVICE) for k, v in inputs.items()}
with torch.no_grad():
outputs = model(**inputs)
upsampled_logits = F.interpolate(
outputs.logits,
size=image.size[::-1], # PIL gives (W, H), torch expects (H, W)
mode="bilinear",
align_corners=False
)
pred = upsampled_logits.argmax(dim=1)[0].detach().cpu().numpy()
color_mask = make_color_mask(pred)
overlay = make_overlay(original_np, color_mask, alpha=0.5)
summary = make_class_summary(pred)
return Image.fromarray(color_mask), Image.fromarray(overlay), summary
title = "SegFormer Semantic Segmentation Demo"
description = (
"Upload an outdoor scene image to generate a semantic segmentation mask and overlay. "
"Classes: background, footpath, grass, road, water."
)
article = """
### Notes
- The model is loaded from the Hugging Face Hub using `MODEL_ID`.
- To use your own model in this Space, set `MODEL_ID` in the Space settings or replace the default value in `app.py`.
- Later you can add your public Space link to the model README.
"""
demo = gr.Interface(
fn=predict,
inputs=gr.Image(type="pil", label="Upload Image"),
outputs=[
gr.Image(type="pil", label="Segmentation Mask"),
gr.Image(type="pil", label="Overlay"),
gr.Markdown(label="Class Summary"),
],
title=title,
description=description,
examples=[
# Add your sample images later, for example:
# ["assets/sample1.jpg"],
# ["assets/sample2.jpg"],
],
)
if __name__ == "__main__":
demo.launch() |