upload

app.py

@@ -79,8 +79,8 @@ def process_image(
         show_progress_bar=False,
     )
 
-    depth_pred = pipe_out.depth_np
-    depth_colored = pipe_out.depth_colored
+    depth_pred = pipe_out.pred_np
+    depth_colored = pipe_out.pred_colored
     depth_16bit = (depth_pred * 65535.0).astype(np.uint16)
 
     np.save(path_out_fp32, depth_pred)
@@ -266,7 +266,7 @@ def main():
 
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     
-    dtype = torch.float16
+    dtype = torch.float32
     
     vae = AutoencoderKL.from_pretrained("guangkaixu/GenPercept", subfolder='vae').to(dtype)
     unet_depth_v1 = UNet2DConditionModel.from_pretrained('guangkaixu/GenPercept', subfolder="unet_depth_v1").to(dtype)

pipeline_genpercept.py

@@ -148,14 +148,10 @@ class GenPerceptPipeline(DiffusionPipeline):
             # Normalize rgb values
             rgb = np.transpose(image, (2, 0, 1))  # [H, W, rgb] -> [rgb, H, W]
             rgb_norm = rgb / 255.0 * 2.0 - 1.0
-            rgb_norm = torch.from_numpy(rgb_norm).to(self.dtype)
+            rgb_norm = torch.from_numpy(rgb_norm).to(self.unet.dtype)
             rgb_norm = rgb_norm[None].to(device)
             assert rgb_norm.min() >= -1.0 and rgb_norm.max() <= 1.0
             bs_imgs = 1
-        
-        print('rgb_norm :', rgb_norm.dtype)
-        print('unet :', self.unet.dtype)
-        print('vae :', self.vae.dtype)
 
         # ----------------- Predicting depth -----------------