update metric3d ViT.giant2 model

depth.py

@@ -14,16 +14,13 @@
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description='Running Metric3D')
     parser.add_argument("--images", help='dir for image files', type=str, required=True)
-    parser.add_argument("--focal", help='focal length', type=float, default=None)
-    parser.add_argument("--calib", help='calibration file, overwrite focal', type=str, default=None)
+    parser.add_argument("--calib", help='calibration file, overwrite focal', type=str, required=True)
     parser.add_argument("--out", help='dir for output depth', type=str, default='')
-    parser.add_argument("--ckpt", type=str, default='./weights/metric_depth_vit_large_800k.pth', help='checkpoint file')
-    parser.add_argument("--model-name", type=str, default='v2-L', choices=['v2-L', 'v2-S'], help='model type')
+    parser.add_argument("--ckpt", type=str, default='./weights/metric_depth_vit_giant2_800k.pth', help='checkpoint file')
+    parser.add_argument("--model-name", type=str, default='v2-g', choices=['v2-L', 'v2-S', 'v2-g'], help='model type')
     args = parser.parse_args()
 
-    if args.calib:
-        calib = np.loadtxt(args.calib)
-        args.focal = (abs(calib[0]) + abs(calib[1])) / 2
+    calib = np.loadtxt(args.calib)
 
     metric = Metric(
         checkpoint=args.ckpt,
@@ -38,7 +35,7 @@
     os.makedirs(color_dir, exist_ok=True)
 
     for image in tqdm(images):
-        depth = metric(image, args.focal)
+        depth = metric(rgb_image=image, calib=calib)
 
         # save orignal depth
         np.save(str(out_dir / f'{image.stem}.npy'), depth)

modules/metric.py

@@ -26,8 +26,8 @@ class Metric3D:
 
     def __init__(
         self,
-        checkpoint: Union[str, Path] = './weights/metric_depth_vit_giant2_800k.pth',
-        model_name: str = 'v2-g',
+        checkpoint: Union[str, Path] = './weights/metric_depth_vit_large_800k.pth',
+        model_name: str = 'v2-L',
     ) -> None:
         checkpoint = Path(checkpoint).resolve()
         cfg:Config = self._load_config_(model_name, checkpoint)
@@ -44,7 +44,7 @@ def __init__(
     def __call__(
         self,
         rgb_image: Union[np.ndarray, Image.Image, str, Path],
-        focal: Optional[float] = None,
+        calib: Union[str, Path, np.ndarray]
     ) -> np.ndarray:
         # read image
         if isinstance(rgb_image, (str, Path)):
@@ -53,9 +53,9 @@ def __call__(
             rgb_image = np.array(rgb_image)
         # get intrinsic
         h, w = rgb_image.shape[:2]
-        if focal is None:
-            focal = np.max([h, w])
-        intrinsic = [focal, focal, w/2, h/2]
+        if isinstance(calib, (str, Path)):
+            calib = np.loadtxt(calib)
+        intrinsic = calib[:4]
         # transform image
         rgb_input, cam_models_stacks, pad, label_scale_factor = \
             transform_test_data_scalecano(rgb_image, intrinsic, self.cfg_.data_basic)
@@ -82,11 +82,12 @@ def _load_config_(
         checkpoint: Union[str, Path],
     ) -> Config:
         config_path = metric3d_path / 'mono/configs/HourglassDecoder'
-        assert model_name in ['v2-L', 'v2-S'], f"Model {model_name} not supported"
+        assert model_name in ['v2-L', 'v2-S', 'v2-g'], f"Model {model_name} not supported"
         # load config file
         cfg = Config.fromfile(
             str(config_path / 'vit.raft5.large.py') if model_name == 'v2-L' 
-            else str(config_path / 'vit.raft5.small.py')
+            else str(config_path / 'vit.raft5.small.py') if model_name == 'v2-S' 
+            else str(config_path / 'vit.raft5.giant2.py')
         )
         cfg.load_from = str(checkpoint)
         # load data info

scripts/test_seq.py

@@ -29,7 +29,7 @@
     parser.add_argument("--global-ba-frontend", type=int, help="frequency to run global ba on frontend", default=0)
 
     parser.add_argument("--metric-ckpt", type=str, help='checkpoint file', default='./weights/metric_depth_vit_large_800k.pth')
-    parser.add_argument("--metric-model", type=str, help='model type', default='v2-L', choices=['v2-L', 'v2-S'])
+    parser.add_argument("--metric-model", type=str, help='model type', default='v2-L', choices=['v2-L', 'v2-S', 'v2-g'])
     parser.add_argument("--droid-ckpt", type=str, help="checkpoint file", default='./weights/droid.pth')
 
     args = parser.parse_args()
@@ -57,7 +57,6 @@
     calib = np.loadtxt(calib_file)
     intr = calib[:4]
     distort = calib[4:] if len(calib) > 4 else None
-    fxy = (intr[0] + intr[1]) / 2
 
     # metric 3d ###############################################################
     metric = Metric(
@@ -70,7 +69,7 @@
         if os.path.exists(str(depth_dir / f'{image.stem}.npy')) and not args.overwrite:
             continue
 
-        depth = metric(image, fxy)
+        depth = metric(image, intr)
         # save orignal depth
         np.save(str(depth_dir / f'{image.stem}.npy'), depth)
         # save colormap

scripts/test_video.py

@@ -30,7 +30,7 @@
     parser.add_argument("--global-ba-frontend", type=int, help="frequency to run global ba on frontend", default=0)
 
     parser.add_argument("--metric-ckpt", type=str, help='checkpoint file', default='./weights/metric_depth_vit_large_800k.pth')
-    parser.add_argument("--metric-model", type=str, help='model type', default='v2-L', choices=['v2-L', 'v2-S'])
+    parser.add_argument("--metric-model", type=str, help='model type', default='v2-L', choices=['v2-L', 'v2-S', 'v2-g'])
     parser.add_argument("--droid-ckpt", type=str, help="checkpoint file", default='./weights/droid.pth')
 
     args = parser.parse_args()
@@ -62,7 +62,6 @@
     calib = np.loadtxt(str(calib_file))
     intr = calib[:4]
     distort = calib[4:] if len(calib) > 4 else None
-    fxy = (intr[0] + intr[1]) / 2
 
     # video sample ###############################################################
     sample_from_video(
@@ -83,7 +82,7 @@
         if os.path.exists(str(depth_dir / f'{image.stem}.npy')) and not args.overwrite:
             continue
 
-        depth = metric(image, fxy)
+        depth = metric(image, intr)
         # save orignal depth
         np.save(str(depth_dir / f'{image.stem}.npy'), depth)
         # save colormap