Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,4 @@
|
|
|
|
1 |
import torch
|
2 |
import gradio as gr
|
3 |
import numpy as np
|
@@ -5,16 +6,18 @@ from PIL import Image
|
|
5 |
import torchvision.transforms.functional as TF
|
6 |
from matplotlib import colormaps
|
7 |
from transformers import AutoModel
|
|
|
8 |
|
9 |
# ----------------------------
|
10 |
# Configuration
|
11 |
# ----------------------------
|
12 |
# The model will be downloaded from the Hugging Face Hub
|
13 |
-
|
|
|
14 |
PATCH_SIZE = 16
|
15 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
16 |
|
17 |
-
# Normalization constants
|
18 |
IMAGENET_MEAN = (0.485, 0.456, 0.406)
|
19 |
IMAGENET_STD = (0.229, 0.224, 0.225)
|
20 |
|
@@ -25,14 +28,17 @@ def load_model_from_hub():
|
|
25 |
"""Loads the DINOv3 model from the Hugging Face Hub."""
|
26 |
print(f"Loading model '{MODEL_ID}' from Hugging Face Hub...")
|
27 |
try:
|
28 |
-
model
|
|
|
|
|
|
|
29 |
model.to(DEVICE).eval()
|
30 |
print(f"β
Model loaded successfully on device: {DEVICE}")
|
31 |
return model
|
32 |
except Exception as e:
|
33 |
print(f"β Failed to load model: {e}")
|
34 |
-
|
35 |
-
|
36 |
|
37 |
# Load the model globally when the app starts
|
38 |
model = load_model_from_hub()
|
@@ -79,7 +85,7 @@ def generate_pca_visuals(
|
|
79 |
):
|
80 |
"""Main function to generate PCA visuals."""
|
81 |
if model is None:
|
82 |
-
raise gr.Error("DINOv3 model
|
83 |
if image_pil is None:
|
84 |
return None, None, "Please upload an image and click Generate.", None, None
|
85 |
|
@@ -94,20 +100,24 @@ def generate_pca_visuals(
|
|
94 |
# 2. Feature Extraction
|
95 |
progress(0.5, desc="π¦ Extracting features with DINOv3...")
|
96 |
outputs = model(t_norm)
|
97 |
-
|
98 |
-
|
|
|
|
|
|
|
|
|
99 |
|
100 |
# 3. PCA Calculation
|
101 |
progress(0.8, desc="π¬ Performing PCA...")
|
102 |
X_centered = patch_embeddings.float() - patch_embeddings.float().mean(0, keepdim=True)
|
103 |
U, S, V = torch.pca_lowrank(X_centered, q=3, center=False)
|
104 |
|
105 |
-
# Stabilize the signs of the eigenvectors for deterministic output
|
|
|
106 |
for i in range(V.shape[1]):
|
107 |
max_abs_idx = torch.argmax(torch.abs(V[:, i]))
|
108 |
if V[max_abs_idx, i] < 0:
|
109 |
V[:, i] *= -1
|
110 |
-
|
111 |
scores = X_centered @ V[:, :3]
|
112 |
|
113 |
# 4. Explained Variance
|
@@ -121,8 +131,10 @@ def generate_pca_visuals(
|
|
121 |
)
|
122 |
|
123 |
# 5. Create Visualizations
|
|
|
124 |
pc1_map = scores[:, 0].reshape(Hp, Wp).cpu().numpy()
|
125 |
pc1_image_raw = colorize(pc1_map, cmap_name)
|
|
|
126 |
pc_rgb_map = scores.reshape(Hp, Wp, 3).cpu().numpy()
|
127 |
min_vals = pc_rgb_map.reshape(-1, 3).min(axis=0)
|
128 |
max_vals = pc_rgb_map.reshape(-1, 3).max(axis=0)
|
@@ -137,7 +149,6 @@ def generate_pca_visuals(
|
|
137 |
progress(1.0, desc="β
Done!")
|
138 |
return pc1_image_smooth, pc_rgb_image_smooth, variance_text, blended_image, original_processed_image
|
139 |
|
140 |
-
|
141 |
# ----------------------------
|
142 |
# Gradio Interface
|
143 |
# ----------------------------
|
@@ -152,7 +163,8 @@ with gr.Blocks(theme=gr.themes.Soft(), title="DINOv3 PCA Explorer") as demo:
|
|
152 |
|
153 |
with gr.Row():
|
154 |
with gr.Column(scale=2):
|
155 |
-
|
|
|
156 |
|
157 |
with gr.Accordion("βοΈ Visualization Controls", open=True):
|
158 |
resolution_slider = gr.Slider(
|
|
|
1 |
+
# app.py
|
2 |
import torch
|
3 |
import gradio as gr
|
4 |
import numpy as np
|
|
|
6 |
import torchvision.transforms.functional as TF
|
7 |
from matplotlib import colormaps
|
8 |
from transformers import AutoModel
|
9 |
+
import os
|
10 |
|
11 |
# ----------------------------
|
12 |
# Configuration
|
13 |
# ----------------------------
|
14 |
# The model will be downloaded from the Hugging Face Hub
|
15 |
+
# Using the specific revision that works well with transformers AutoModel
|
16 |
+
MODEL_ID = "facebook/dinov3-vith16plus"
|
17 |
PATCH_SIZE = 16
|
18 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
19 |
|
20 |
+
# Normalization constants (standard for ImageNet)
|
21 |
IMAGENET_MEAN = (0.485, 0.456, 0.406)
|
22 |
IMAGENET_STD = (0.229, 0.224, 0.225)
|
23 |
|
|
|
28 |
"""Loads the DINOv3 model from the Hugging Face Hub."""
|
29 |
print(f"Loading model '{MODEL_ID}' from Hugging Face Hub...")
|
30 |
try:
|
31 |
+
# Use your HF token if the model is gated
|
32 |
+
# You can set this as a secret in your Hugging Face Space settings
|
33 |
+
token = os.environ.get("HF_TOKEN")
|
34 |
+
model = AutoModel.from_pretrained(MODEL_ID, token=token, trust_remote_code=True)
|
35 |
model.to(DEVICE).eval()
|
36 |
print(f"β
Model loaded successfully on device: {DEVICE}")
|
37 |
return model
|
38 |
except Exception as e:
|
39 |
print(f"β Failed to load model: {e}")
|
40 |
+
# This will display an error message in the Gradio interface
|
41 |
+
raise gr.Error(f"Could not load model from Hub. If it's a gated model, ensure you have access and have set your HF_TOKEN secret in the Space settings. Error: {e}")
|
42 |
|
43 |
# Load the model globally when the app starts
|
44 |
model = load_model_from_hub()
|
|
|
85 |
):
|
86 |
"""Main function to generate PCA visuals."""
|
87 |
if model is None:
|
88 |
+
raise gr.Error("DINOv3 model is not available. Check the startup logs.")
|
89 |
if image_pil is None:
|
90 |
return None, None, "Please upload an image and click Generate.", None, None
|
91 |
|
|
|
100 |
# 2. Feature Extraction
|
101 |
progress(0.5, desc="π¦ Extracting features with DINOv3...")
|
102 |
outputs = model(t_norm)
|
103 |
+
|
104 |
+
# π‘ FIX: The model output includes a [CLS] token AND 4 register tokens.
|
105 |
+
# We must skip all of them (total 5) to get only the patch embeddings.
|
106 |
+
# The original code only skipped 1, causing the size mismatch.
|
107 |
+
n_special_tokens = 5 # 1 [CLS] token + 4 register tokens
|
108 |
+
patch_embeddings = outputs.last_hidden_state.squeeze(0)[n_special_tokens:, :]
|
109 |
|
110 |
# 3. PCA Calculation
|
111 |
progress(0.8, desc="π¬ Performing PCA...")
|
112 |
X_centered = patch_embeddings.float() - patch_embeddings.float().mean(0, keepdim=True)
|
113 |
U, S, V = torch.pca_lowrank(X_centered, q=3, center=False)
|
114 |
|
115 |
+
# π‘ IMPROVEMENT: Stabilize the signs of the eigenvectors for deterministic output.
|
116 |
+
# This prevents the colors from randomly inverting on different runs.
|
117 |
for i in range(V.shape[1]):
|
118 |
max_abs_idx = torch.argmax(torch.abs(V[:, i]))
|
119 |
if V[max_abs_idx, i] < 0:
|
120 |
V[:, i] *= -1
|
|
|
121 |
scores = X_centered @ V[:, :3]
|
122 |
|
123 |
# 4. Explained Variance
|
|
|
131 |
)
|
132 |
|
133 |
# 5. Create Visualizations
|
134 |
+
# This part should now work correctly as `scores` has the right shape (Hp*Wp, 3)
|
135 |
pc1_map = scores[:, 0].reshape(Hp, Wp).cpu().numpy()
|
136 |
pc1_image_raw = colorize(pc1_map, cmap_name)
|
137 |
+
|
138 |
pc_rgb_map = scores.reshape(Hp, Wp, 3).cpu().numpy()
|
139 |
min_vals = pc_rgb_map.reshape(-1, 3).min(axis=0)
|
140 |
max_vals = pc_rgb_map.reshape(-1, 3).max(axis=0)
|
|
|
149 |
progress(1.0, desc="β
Done!")
|
150 |
return pc1_image_smooth, pc_rgb_image_smooth, variance_text, blended_image, original_processed_image
|
151 |
|
|
|
152 |
# ----------------------------
|
153 |
# Gradio Interface
|
154 |
# ----------------------------
|
|
|
163 |
|
164 |
with gr.Row():
|
165 |
with gr.Column(scale=2):
|
166 |
+
# Added a default image URL for convenience
|
167 |
+
input_image = gr.Image(type="pil", label="Upload Image", value="https://images.squarespace-cdn.com/content/v1/607f89e638219e13eee71b1e/1684821560422-SD5V37BAG28BURTLIXUQ/michael-sum-LEpfefQf4rU-unsplash.jpg")
|
168 |
|
169 |
with gr.Accordion("βοΈ Visualization Controls", open=True):
|
170 |
resolution_slider = gr.Slider(
|