Spaces:
Running
Running
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,138 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import torch
|
3 |
+
from neuralop.models import FNO # Make sure this import path matches your environment
|
4 |
+
import matplotlib.pyplot as plt
|
5 |
+
import numpy as np
|
6 |
+
import os # For path handling
|
7 |
+
|
8 |
+
# --- Configuration ---
|
9 |
+
# Set paths relative to the root of your Hugging Face Space repository
|
10 |
+
MODEL_PATH = "fno_ckpt_single_res"
|
11 |
+
DATASET_PATH = "navier_stokes_2d.pt" # Ensure this file is in your repo root
|
12 |
+
|
13 |
+
# --- Global Variables for Model and Data (loaded once) ---
|
14 |
+
MODEL = None
|
15 |
+
FULL_DATASET_X = None # To store all initial conditions
|
16 |
+
|
17 |
+
# --- 1. Model Loading Function ---
|
18 |
+
def load_model():
|
19 |
+
"""Loads the pre-trained FNO model."""
|
20 |
+
global MODEL
|
21 |
+
if MODEL is None:
|
22 |
+
print("Loading FNO model...")
|
23 |
+
try:
|
24 |
+
# Ensure model is loaded to CPU for general compatibility in Spaces
|
25 |
+
MODEL = torch.load(MODEL_PATH, weights_only=False, map_location='cpu')
|
26 |
+
MODEL.eval() # Set to evaluation mode
|
27 |
+
print("Model loaded successfully.")
|
28 |
+
except Exception as e:
|
29 |
+
print(f"Error loading model: {e}")
|
30 |
+
raise gr.Error(f"Failed to load model: {e}")
|
31 |
+
return MODEL
|
32 |
+
|
33 |
+
# --- 2. Dataset Loading Function ---
|
34 |
+
def load_dataset():
|
35 |
+
"""Loads the initial conditions dataset."""
|
36 |
+
global FULL_DATASET_X
|
37 |
+
if FULL_DATASET_X is None:
|
38 |
+
print("Loading dataset...")
|
39 |
+
try:
|
40 |
+
data = torch.load(DATASET_PATH, map_location='cpu')
|
41 |
+
if isinstance(data, dict) and 'x' in data:
|
42 |
+
FULL_DATASET_X = data['x']
|
43 |
+
elif isinstance(data, torch.Tensor):
|
44 |
+
FULL_DATASET_X = data
|
45 |
+
else:
|
46 |
+
raise ValueError("Unknown dataset format or 'x' key missing.")
|
47 |
+
print(f"Dataset loaded. Total samples: {FULL_DATASET_X.shape[0]}")
|
48 |
+
except FileNotFoundError:
|
49 |
+
print(f"Dataset file not found at {DATASET_PATH}")
|
50 |
+
raise gr.Error(f"Dataset file not found. Please ensure '{DATASET_PATH}' is in your Space.")
|
51 |
+
except Exception as e:
|
52 |
+
print(f"Error loading dataset: {e}")
|
53 |
+
raise gr.Error(f"Failed to load dataset: {e}")
|
54 |
+
return FULL_DATASET_X
|
55 |
+
|
56 |
+
# --- 3. Inference Function for Gradio ---
|
57 |
+
def run_inference(sample_index: int):
|
58 |
+
"""
|
59 |
+
Performs inference for a selected sample index from the dataset.
|
60 |
+
Returns two Matplotlib figures: one for input, one for output.
|
61 |
+
"""
|
62 |
+
model = load_model()
|
63 |
+
dataset = load_dataset()
|
64 |
+
|
65 |
+
if not (0 <= sample_index < dataset.shape[0]):
|
66 |
+
raise gr.Error(f"Sample index out of range. Please choose between 0 and {dataset.shape[0]-1}.")
|
67 |
+
|
68 |
+
# Extract single initial condition and add channel dimension
|
69 |
+
# (shape: [1, H, W] -> [1, 1, H, W])
|
70 |
+
single_initial_condition = dataset[sample_index:sample_index+1, :, :].unsqueeze(1)
|
71 |
+
|
72 |
+
print(f"Running inference for sample index {sample_index}...")
|
73 |
+
with torch.no_grad():
|
74 |
+
predicted_solution = model(single_initial_condition)
|
75 |
+
|
76 |
+
# Convert tensors to numpy for plotting
|
77 |
+
input_numpy = single_initial_condition.squeeze().cpu().numpy()
|
78 |
+
output_numpy = predicted_solution.squeeze().cpu().numpy()
|
79 |
+
|
80 |
+
# Create Matplotlib figures
|
81 |
+
fig_input, ax_input = plt.subplots()
|
82 |
+
im_input = ax_input.imshow(input_numpy, cmap='viridis')
|
83 |
+
ax_input.set_title(f"Initial Condition (Sample {sample_index})")
|
84 |
+
fig_input.colorbar(im_input, ax=ax_input, label="Vorticity")
|
85 |
+
plt.close(fig_input) # Close to prevent display issues in Gradio
|
86 |
+
|
87 |
+
fig_output, ax_output = plt.subplots()
|
88 |
+
im_output = ax_output.imshow(output_numpy, cmap='viridis')
|
89 |
+
ax_output.set_title(f"Predicted Solution")
|
90 |
+
fig_output.colorbar(im_output, ax=ax_output, label="Vorticity")
|
91 |
+
plt.close(fig_output) # Close to prevent display issues in Gradio
|
92 |
+
|
93 |
+
return fig_input, fig_output
|
94 |
+
|
95 |
+
# --- Gradio Interface Setup ---
|
96 |
+
with gr.Blocks() as demo:
|
97 |
+
gr.Markdown(
|
98 |
+
"""
|
99 |
+
# Fourier Neural Operator (FNO) for Navier-Stokes Equations
|
100 |
+
Select a sample index from the pre-loaded dataset to see the FNO's prediction
|
101 |
+
of the vorticity field evolution.
|
102 |
+
"""
|
103 |
+
)
|
104 |
+
|
105 |
+
with gr.Row():
|
106 |
+
with gr.Column():
|
107 |
+
sample_input_slider = gr.Slider(
|
108 |
+
minimum=0,
|
109 |
+
maximum=9999, # Assuming 10,000 samples based on your dataset shape
|
110 |
+
value=0,
|
111 |
+
step=1,
|
112 |
+
label="Select Sample Index"
|
113 |
+
)
|
114 |
+
run_button = gr.Button("Generate Solution")
|
115 |
+
with gr.Column():
|
116 |
+
input_image_plot = gr.Plot(label="Selected Initial Condition")
|
117 |
+
output_image_plot = gr.Plot(label="Predicted Solution")
|
118 |
+
|
119 |
+
# Bind the button click to the inference function
|
120 |
+
run_button.click(
|
121 |
+
fn=run_inference,
|
122 |
+
inputs=[sample_input_slider],
|
123 |
+
outputs=[input_image_plot, output_image_plot]
|
124 |
+
)
|
125 |
+
|
126 |
+
# Optional: Load initial data on startup for the first display
|
127 |
+
def load_initial_data_and_predict():
|
128 |
+
# Ensure model and dataset are loaded when the space starts
|
129 |
+
load_model()
|
130 |
+
load_dataset()
|
131 |
+
# Run inference for the default value (index 0)
|
132 |
+
return run_inference(0)
|
133 |
+
|
134 |
+
demo.load(load_initial_data_and_predict, inputs=None, outputs=[input_image_plot, output_image_plot])
|
135 |
+
|
136 |
+
# Launch the Gradio app (only runs when you test locally)
|
137 |
+
if __name__ == "__main__":
|
138 |
+
demo.launch()
|