Commit
·
f71d4bc
1
Parent(s):
7d74b9b
solve problem with input third try
Browse files
app.py
CHANGED
|
@@ -5,28 +5,44 @@ from PIL import Image
|
|
| 5 |
import cv2
|
| 6 |
|
| 7 |
# Загрузка модели
|
| 8 |
-
|
| 9 |
-
midas = torch.hub.load("intel-isl/MiDaS", "MiDaS_small").to(device)
|
| 10 |
midas.eval()
|
| 11 |
|
| 12 |
-
midas_transforms = torch.hub.load("intel-isl/MiDaS", "transforms")
|
| 13 |
transform = midas_transforms.small_transform
|
| 14 |
|
| 15 |
-
|
| 16 |
def predict_depth(image):
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
with torch.no_grad():
|
| 22 |
prediction = midas(input_tensor)
|
| 23 |
prediction = torch.nn.functional.interpolate(
|
| 24 |
prediction.unsqueeze(1),
|
| 25 |
-
size=
|
| 26 |
mode="bicubic",
|
| 27 |
align_corners=False,
|
| 28 |
).squeeze()
|
| 29 |
|
|
|
|
| 30 |
depth_map = prediction.cpu().numpy()
|
| 31 |
depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
|
| 32 |
depth_map = (depth_map * 255).astype(np.uint8)
|
|
@@ -34,13 +50,13 @@ def predict_depth(image):
|
|
| 34 |
|
| 35 |
return depth_img
|
| 36 |
|
| 37 |
-
#
|
| 38 |
iface = gr.Interface(
|
| 39 |
fn=predict_depth,
|
| 40 |
inputs=gr.Image(type="pil"),
|
| 41 |
outputs=gr.Image(type="pil"),
|
| 42 |
title="MiDaS Depth Estimation",
|
| 43 |
-
description="Загрузите изображение
|
| 44 |
)
|
| 45 |
|
| 46 |
if __name__ == "__main__":
|
|
|
|
| 5 |
import cv2
|
| 6 |
|
| 7 |
# Загрузка модели
|
| 8 |
+
midas = torch.hub.load("intel-isl/MiDaS", "MiDaS_small")
|
|
|
|
| 9 |
midas.eval()
|
| 10 |
|
| 11 |
+
midas_transforms = torch.hub.load("intel-isl/MiDaS", "transforms")
|
| 12 |
transform = midas_transforms.small_transform
|
| 13 |
|
|
|
|
| 14 |
def predict_depth(image):
|
| 15 |
+
# ======= 1. Проверка типа входных данных =======
|
| 16 |
+
if isinstance(image, torch.Tensor):
|
| 17 |
+
print(f"Пришёл Tensor с формой: {image.shape}")
|
| 18 |
+
|
| 19 |
+
if len(image.shape) == 4:
|
| 20 |
+
input_tensor = image # уже батч [1, 3, H, W]
|
| 21 |
+
elif len(image.shape) == 3:
|
| 22 |
+
input_tensor = image.unsqueeze(0) # сделаем батч
|
| 23 |
+
else:
|
| 24 |
+
raise ValueError(f"Неожиданный размер Tensor: {image.shape}")
|
| 25 |
+
|
| 26 |
+
else:
|
| 27 |
+
print("Пришёл PIL Image или numpy array")
|
| 28 |
+
# Если пришло обычное изображение (PIL или numpy)
|
| 29 |
+
if not isinstance(image, Image.Image):
|
| 30 |
+
image = Image.fromarray(image)
|
| 31 |
+
img = np.array(image)
|
| 32 |
+
img_rgb = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
|
| 33 |
+
input_tensor = transform(img_rgb).unsqueeze(0)
|
| 34 |
+
|
| 35 |
+
# ======= 2. Предсказание =======
|
| 36 |
with torch.no_grad():
|
| 37 |
prediction = midas(input_tensor)
|
| 38 |
prediction = torch.nn.functional.interpolate(
|
| 39 |
prediction.unsqueeze(1),
|
| 40 |
+
size=(input_tensor.shape[2], input_tensor.shape[3]),
|
| 41 |
mode="bicubic",
|
| 42 |
align_corners=False,
|
| 43 |
).squeeze()
|
| 44 |
|
| 45 |
+
# ======= 3. Нормализация карты глубины =======
|
| 46 |
depth_map = prediction.cpu().numpy()
|
| 47 |
depth_map = (depth_map - depth_map.min()) / (depth_map.max() - depth_map.min())
|
| 48 |
depth_map = (depth_map * 255).astype(np.uint8)
|
|
|
|
| 50 |
|
| 51 |
return depth_img
|
| 52 |
|
| 53 |
+
# Gradio интерфейс
|
| 54 |
iface = gr.Interface(
|
| 55 |
fn=predict_depth,
|
| 56 |
inputs=gr.Image(type="pil"),
|
| 57 |
outputs=gr.Image(type="pil"),
|
| 58 |
title="MiDaS Depth Estimation",
|
| 59 |
+
description="Загрузите изображение или отправьте через API. Получите карту глубины."
|
| 60 |
)
|
| 61 |
|
| 62 |
if __name__ == "__main__":
|