adding camera offsets values

inference/v2v_data/demo.py

@@ -588,84 +588,21 @@ class GetAnchorVideos:
             .to(opts.device)
         )
 
-        camera_x = getattr(opts, "init_dx", 0.0)
-        camera_y = getattr(opts, "init_dy", 0.0)
-        camera_z = getattr(opts, "init_dz", 0.0)
-
-        print("camera_x", camera_x)
-        print("camera_y", camera_y)
-        print("camera_z", camera_z)
-
         c2w_init = (
             torch.tensor(
                 [
-                    [-1.0, 0.0, 0.0, camera_x],
-                    [0.0, 1.0, 0.0, camera_y],
-                    [0.0, 0.0, -1.0, camera_z],
+                    [-1.0, 0.0, 0.0, 0],
+                    [0.0, 1.0, 0.0, 0],
+                    [0.0, 0.0, -1.0, 0],
                     [0.0, 0.0, 0.0, 1.0],
                 ]
             ).to(opts.device).unsqueeze(0)
         )
-            
-        theta = torch.deg2rad(torch.tensor(0)).to(opts.device)
-        sin_value_x = torch.sin(theta)
-        cos_value_x = torch.cos(theta)
-        rot_mat_x = (
-            torch.tensor(
-                [
-                    [1, 0, 0, 0],
-                    [0, cos_value_x, -sin_value_x, 0],
-                    [0, sin_value_x, cos_value_x, 0],
-                    [0, 0, 0, 1],
-                ]
-            )
-            .unsqueeze(0)
-            .repeat(c2w_init.shape[0], 1, 1)
-            .to(opts.device)
-        )
-
-        phi = torch.deg2rad(torch.tensor(0)).to(opts.device)
-        sin_value_y = torch.sin(phi)
-        cos_value_y = torch.cos(phi)
-        rot_mat_y = (
-            torch.tensor(
-                [
-                    [cos_value_y, 0, sin_value_y, 0],
-                    [0, 1, 0, 0],
-                    [-sin_value_y, 0, cos_value_y, 0],
-                    [0, 0, 0, 1],
-                ]
-            )
-            .unsqueeze(0)
-            .repeat(c2w_init.shape[0], 1, 1)
-            .to(opts.device)
-        )
-
-        c2w_init = torch.matmul(rot_mat_x, c2w_init)
-        c2w_init = torch.matmul(rot_mat_y, c2w_init)
-        # c2ws[:, 2, 3] -= r
-        # if x is not None:
-        #     c2ws[:, 1, 3] += y
-        # if y is not None:
-        #     c2ws[:, 0, 3] -= x
-
-        # c2w_init = (
-        #     torch.tensor(
-        #         [
-        #             [-1.0, 0.0, 0.0, 0.0],
-        #             [0.0, 1.0, 0.0, 0.0],
-        #             [0.0, 0.0, -1.0, 0.0],
-        #             [0.0, 0.0, 0.0, 1.0],
-        #         ]
-        #     )
-        #     .to(opts.device)
-        #     .unsqueeze(0)
-        # )
         
         if opts.camera == 'target':
             dtheta, dphi, dr, dx, dy = opts.target_pose
             poses = generate_traj_specified(
-                c2w_init, dtheta, dphi, dr * radius, dx, dy, num_frames, opts.device
+                c2w_init, dtheta, dphi, dr * radius, dx, dy, num_frames, opts.device, opts
             )
         elif opts.camera == 'target_fast':
             dtheta, dphi, dr, dx, dy = opts.target_pose

inference/v2v_data/models/utils.py

@@ -180,14 +180,20 @@ def sphere2pose(c2ws_input, theta, phi, r, device, x=None, y=None):
     return c2ws
 
 
-def generate_traj_specified(c2ws_anchor, theta, phi, d_r, d_x, d_y, frame, device):
+def generate_traj_specified(c2ws_anchor, theta, phi, d_r, d_x, d_y, frame, device, opts):
     # Initialize a camera.
+
+    camera_x = getattr(opts, "init_dx", 0.0)
+    camera_y = getattr(opts, "init_dy", 0.0)
+    camera_z = getattr(opts, "init_dz", 0.0)
+
     thetas = np.linspace(0, theta, frame)
     phis = np.linspace(0, phi, frame)
-    rs = np.linspace(0, d_r, frame)
-    xs = np.linspace(0, d_x, frame)
-    ys = np.linspace(0, d_y, frame)
-    
+
+    rs = np.linspace(0, d_r, frame)+camera_z
+    xs = np.linspace(0, d_x, frame)+camera_x
+    ys = np.linspace(0, d_y, frame)+camera_y
+    # manipulate the above ones 
     c2ws_list = []
     for th, ph, r, x, y in zip(thetas, phis, rs, xs, ys):
         c2w_new = sphere2pose(