Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -2,12 +2,12 @@ import gradio as gr
|
|
| 2 |
import torch
|
| 3 |
import numpy as np
|
| 4 |
from PIL import Image
|
| 5 |
-
from huggingface_hub import hf_hub_download
|
| 6 |
import matplotlib.pyplot as plt
|
| 7 |
-
|
|
|
|
| 8 |
from depth_anything_v2.dpt import DepthAnythingV2
|
| 9 |
|
| 10 |
-
#
|
| 11 |
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 12 |
model_configs = {
|
| 13 |
'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
|
|
@@ -23,21 +23,28 @@ state_dict = torch.load(model_path, map_location="cpu")
|
|
| 23 |
model.load_state_dict(state_dict)
|
| 24 |
model = model.to(DEVICE).eval()
|
| 25 |
|
| 26 |
-
# Use a matplotlib colormap
|
| 27 |
CMAP = plt.get_cmap('Spectral_r')
|
| 28 |
|
| 29 |
def infer(image: np.ndarray):
|
| 30 |
-
#
|
| 31 |
with torch.no_grad():
|
| 32 |
depth = model.infer_image(image[:, :, ::-1])
|
| 33 |
-
|
|
|
|
| 34 |
depth_norm = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
|
| 35 |
depth_norm = depth_norm.astype(np.uint8)
|
| 36 |
gray = Image.fromarray(depth_norm)
|
| 37 |
-
|
|
|
|
| 38 |
colored = (CMAP(depth_norm)[:, :, :3] * 255).astype(np.uint8)
|
| 39 |
color = Image.fromarray(colored)
|
| 40 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 41 |
|
| 42 |
iface = gr.Interface(
|
| 43 |
fn=infer,
|
|
@@ -45,9 +52,10 @@ iface = gr.Interface(
|
|
| 45 |
outputs=[
|
| 46 |
gr.Image(label="Grayscale Depth"),
|
| 47 |
gr.Image(label="Colored Depth"),
|
|
|
|
| 48 |
],
|
| 49 |
-
title="Depth Anything V2 (
|
| 50 |
-
description="Upload an image
|
| 51 |
)
|
| 52 |
|
| 53 |
iface.launch()
|
|
|
|
| 2 |
import torch
|
| 3 |
import numpy as np
|
| 4 |
from PIL import Image
|
|
|
|
| 5 |
import matplotlib.pyplot as plt
|
| 6 |
+
import cv2
|
| 7 |
+
from huggingface_hub import hf_hub_download
|
| 8 |
from depth_anything_v2.dpt import DepthAnythingV2
|
| 9 |
|
| 10 |
+
# Model loading (as before)
|
| 11 |
DEVICE = 'cuda' if torch.cuda.is_available() else 'cpu'
|
| 12 |
model_configs = {
|
| 13 |
'vitl': {'encoder': 'vitl', 'features': 256, 'out_channels': [256, 512, 1024, 1024]},
|
|
|
|
| 23 |
model.load_state_dict(state_dict)
|
| 24 |
model = model.to(DEVICE).eval()
|
| 25 |
|
|
|
|
| 26 |
CMAP = plt.get_cmap('Spectral_r')
|
| 27 |
|
| 28 |
def infer(image: np.ndarray):
|
| 29 |
+
# Run depth model
|
| 30 |
with torch.no_grad():
|
| 31 |
depth = model.infer_image(image[:, :, ::-1])
|
| 32 |
+
|
| 33 |
+
# Grayscale map (normalize to 0..255)
|
| 34 |
depth_norm = (depth - depth.min()) / (depth.max() - depth.min()) * 255.0
|
| 35 |
depth_norm = depth_norm.astype(np.uint8)
|
| 36 |
gray = Image.fromarray(depth_norm)
|
| 37 |
+
|
| 38 |
+
# Colored map
|
| 39 |
colored = (CMAP(depth_norm)[:, :, :3] * 255).astype(np.uint8)
|
| 40 |
color = Image.fromarray(colored)
|
| 41 |
+
|
| 42 |
+
# Edge map using Canny on the original image (convert to grayscale first)
|
| 43 |
+
image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) if image.shape[2] == 3 else image
|
| 44 |
+
edges = cv2.Canny(image_gray, 100, 200) # threshold1/2 can be tuned
|
| 45 |
+
edge_img = Image.fromarray(edges)
|
| 46 |
+
|
| 47 |
+
return gray, color, edge_img
|
| 48 |
|
| 49 |
iface = gr.Interface(
|
| 50 |
fn=infer,
|
|
|
|
| 52 |
outputs=[
|
| 53 |
gr.Image(label="Grayscale Depth"),
|
| 54 |
gr.Image(label="Colored Depth"),
|
| 55 |
+
gr.Image(label="Canny Edge Map"),
|
| 56 |
],
|
| 57 |
+
title="Depth Anything V2 (with Colored Output + Canny Edges)",
|
| 58 |
+
description="Upload an image to get depth (gray, color), plus Canny edge map."
|
| 59 |
)
|
| 60 |
|
| 61 |
iface.launch()
|