From f96d66afedd008976fdfc18fb50b4984bb454195 Mon Sep 17 00:00:00 2001
From: Gengshan Yang <y.gengshan@gmail.com>
Date: Thu, 30 May 2024 03:07:25 -0400
Subject: [PATCH] fix manual camera

---
 preprocess/scripts/manual_cameras.py | 19 +++++++++++++------
 scripts/run_preprocess.py            |  2 +-
 2 files changed, 14 insertions(+), 7 deletions(-)

diff --git a/preprocess/scripts/manual_cameras.py b/preprocess/scripts/manual_cameras.py
index 53451e9..ea4e71c 100644
--- a/preprocess/scripts/manual_cameras.py
+++ b/preprocess/scripts/manual_cameras.py
@@ -318,9 +318,7 @@ def load_fig(track):
 
 def caminfo_to_rotation(track):
     caminfo = track.fig.layout.scene.camera
-    final_R = np.zeros((4, 4))
-    final_R[2, 3] = 3
-    final_R[3, 3] = 1
+    final_R = np.eye(4)
     eye = np.asarray([caminfo.eye.x, caminfo.eye.y, caminfo.eye.z])
     center = np.asarray([caminfo.center.x, caminfo.center.y, caminfo.center.z])
     up = np.asarray([caminfo.up.x, caminfo.up.y, caminfo.up.z])
@@ -329,9 +327,18 @@ def caminfo_to_rotation(track):
     s = np.cross(L, up)
     s = s / np.linalg.norm(s)
     u = np.cross(s, L)
-    cam_R = np.array([s, u, -L])
-    final_R[0:3, 0:3] = cam_R @ track.R
-    final_R[0:3, 0:3] = final_R[0:3, 0:3].T  # field to camera
+    final_R[:3, :3] = np.array([s, u, -L]).T # camera x/y/z coordinate
+    final_R[:3, 3] = eye # camera center
+    final_R = np.linalg.inv(final_R) # object to camera
+
+    # user input adjustment
+    final_R[:3, :3] = final_R[:3, :3] @ track.R
+
+    # gl to cv camera coordinate
+    final_R[1:3, :] *= -1
+    # # gl to cv object coordinate
+    # final_R[:3,1:3] *= -1
+
     # debug_format(cam_R @ track.R)
 
     track.se3_dict[int(track.curr_frame)] = final_R.tolist()
diff --git a/scripts/run_preprocess.py b/scripts/run_preprocess.py
index b06ba07..aa5f339 100644
--- a/scripts/run_preprocess.py
+++ b/scripts/run_preprocess.py
@@ -153,7 +153,7 @@ def run_extract_priors(seqname, outdir, obj_class_cam):
         from preprocess.scripts.manual_cameras import manual_camera_interface
 
         mesh_path = "database/mesh-templates/cat-pikachu-remeshed.obj"
-        manual_camera_interface(vidname, use_manual_segment, mesh_path)
+        manual_camera_interface(vidname, mesh_path)
 
     # extract flow/depth/camera/etc
     gpu_map(run_extract_priors, prior_args, gpus=gpulist)