Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,18 +1,18 @@
|
|
| 1 |
import subprocess
|
| 2 |
-
def install_mmcv():
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
install_mmcv()
|
| 12 |
import mmcv
|
| 13 |
import gradio as gr
|
| 14 |
import numpy as np
|
| 15 |
-
|
| 16 |
import torch
|
| 17 |
import os
|
| 18 |
import cv2
|
|
@@ -40,7 +40,7 @@ model_ckpt_map = {
|
|
| 40 |
}
|
| 41 |
|
| 42 |
# load model
|
| 43 |
-
|
| 44 |
def load_model(model_type: str):
|
| 45 |
global net
|
| 46 |
ckpt_path = model_ckpt_map[model_type]
|
|
@@ -59,7 +59,7 @@ load_model("RGB-D-D")
|
|
| 59 |
|
| 60 |
|
| 61 |
# data process
|
| 62 |
-
|
| 63 |
def preprocess_inputs(rgb_image: Image.Image, lr_depth: Image.Image):
|
| 64 |
image = np.array(rgb_image.convert("RGB")).astype(np.float32)
|
| 65 |
h, w, _ = image.shape
|
|
@@ -81,7 +81,7 @@ def preprocess_inputs(rgb_image: Image.Image, lr_depth: Image.Image):
|
|
| 81 |
|
| 82 |
|
| 83 |
# model inference
|
| 84 |
-
|
| 85 |
@torch.no_grad()
|
| 86 |
def infer(rgb_image: Image.Image, lr_depth: Image.Image, model_type: str):
|
| 87 |
load_model(model_type) # reset weight
|
|
|
|
| 1 |
import subprocess
|
| 2 |
+
# def install_mmcv():
|
| 3 |
+
# try:
|
| 4 |
+
# subprocess.run([
|
| 5 |
+
# "pip", "install", "mmcv-full==1.7.2",
|
| 6 |
+
# "-f", "https://download.openmmlab.com/mmcv/dist/cu121/torch2.1.0/"
|
| 7 |
+
# ], check=True)
|
| 8 |
+
# except subprocess.CalledProcessError as e:
|
| 9 |
+
# print("Failed to install mmcv-full:", e)
|
| 10 |
+
|
| 11 |
+
# install_mmcv()
|
| 12 |
import mmcv
|
| 13 |
import gradio as gr
|
| 14 |
import numpy as np
|
| 15 |
+
import spaces
|
| 16 |
import torch
|
| 17 |
import os
|
| 18 |
import cv2
|
|
|
|
| 40 |
}
|
| 41 |
|
| 42 |
# load model
|
| 43 |
+
@spaces.GPU
|
| 44 |
def load_model(model_type: str):
|
| 45 |
global net
|
| 46 |
ckpt_path = model_ckpt_map[model_type]
|
|
|
|
| 59 |
|
| 60 |
|
| 61 |
# data process
|
| 62 |
+
@spaces.GPU
|
| 63 |
def preprocess_inputs(rgb_image: Image.Image, lr_depth: Image.Image):
|
| 64 |
image = np.array(rgb_image.convert("RGB")).astype(np.float32)
|
| 65 |
h, w, _ = image.shape
|
|
|
|
| 81 |
|
| 82 |
|
| 83 |
# model inference
|
| 84 |
+
@spaces.GPU
|
| 85 |
@torch.no_grad()
|
| 86 |
def infer(rgb_image: Image.Image, lr_depth: Image.Image, model_type: str):
|
| 87 |
load_model(model_type) # reset weight
|