The generated video is not from the same viewpoint as my depth video.

[kaihekaihe]

I have an accident video recorded by a roadside camera. I used Viser to apply an offset to the camera position. In this setup, However, I found that the generated video is not from the same viewpoint as my depth video.

The out out video is

joker.mp4

The meta json file is like: cat ./example/in_the_wild/meta3.json
{
"test_datasets": [
{
"exo_path": "./example/in_the_wild/videos/joker/exo.mp4",
"ego_prior_path": "./example/in_the_wild/videos/joker/ego_Prior3.mp4",
"prompt":"[Exo view]\nScene Overview:\nThe scene is set on a street in front of a hospital, indicated by the large "EMERGENCY" sign visible in the background. The ground is asphalt, marked with yellow lines, and appears to have debris scattered across it, including small rocks and larger pieces of material. To the left, a gray sedan is parked, partially obscured by smoke. In the background, an ambulance is visible, and smoke billows from the building, suggesting a fire or explosion. The building itself has large glass windows and a beige exterior. A red fire hydrant is positioned on the right side of the frame, and the area is shrouded in a haze of smoke, reducing visibility.\n\nAction Analysis:\nThe sequence begins with a figure standing in the middle of the street, wearing a white uniform with red trim. The figure's hands are initially positioned in front of them, possibly holding an object that is not clearly visible. In the second frame, the figure raises a small metallic object, which appears to be a lighter or similar item, with the right hand while the left hand remains open and slightly raised. The third frame captures a moment of ignition, as flames erupt in the background, indicating an explosion or fire. The figure's hands are now in a more dynamic position, with the right hand holding the lighter and the left hand moving as if reacting to the sudden event. In the final frame, the figure turns away from the explosion, with the left arm extended outward and the right hand still gripping the lighter. The transition from a static pose to a reactionary movement illustrates a shift from calm to urgency, as the figure responds to the unfolding chaos.\n\n[Ego view]\nScene Overview:\nFrom the inferred first-person perspective, the environment appears chaotic and filled with smoke. The "EMERGENCY" sign looms in the background, indicating proximity to a hospital. The ground is uneven with debris scattered around, and the asphalt is hot underfoot. The figure's uniform is white with red trim, and the hands are visible, one holding a small metallic lighter. The view is partially obscured by smoke, creating a sense of urgency and confusion.\n\nAction Analysis:\nThe sequence begins with a figure standing in the middle of the street, wearing a white uniform with red trim. The figure's hands are initially positioned in front of them, possibly holding an object that is not clearly visible. In the second frame, the figure raises a small metallic object, which appears to be a lighter or similar item, with the right hand while the left hand remains open and slightly raised. The third frame captures a moment of ignition, as flames erupt in the background, indicating an explosion or fire. The figure's hands are now in a more dynamic position, with the right hand holding the lighter and the left hand moving as if reacting to the sudden event. In the final frame, the figure turns away from the explosion, with the left arm extended outward and the right hand still gripping the lighter. The transition from a static pose to a reactionary movement illustrates a shift from calm to urgency, as the figure responds to the unfolding chaos.",
"camera_intrinsics": [
[
634.47327,
0.0,
392.0
],
[
0.0,
634.4733,
224.0
],
[
0.0,
0.0,
1.0
]
],
"camera_extrinsics": [
[
1.0,
0.0,
0.0,
0.0
],
[
0.0,
1.0,
0.0,
0.0
],
[
0.0,
0.0,
1.0,
0.0
]
],
"ego_intrinsics": [
[
150.0,
0.0,
255.5
],
[
0.0,
150.0,
255.5
],
[
0.0,
0.0,
1.0
]
],
"ego_extrinsics": [
[
[ -0.0000, 0.9916, -0.1294, 1.2776],
[ -0.7802, -0.0809, -0.6203, 2.6248],
[ -0.6255, 0.1009, 0.7736, 0.4278]
],
[
[ -0.0000, 0.9916, -0.1294, 1.2776],
[ -0.7802, -0.0809, -0.6203, 2.6248],
[ -0.6255, 0.1009, 0.7736, 0.4278]
],
[
[ -0.0000, 0.9916, -0.1294, 1.2776],
[ -0.7802, -0.0809, -0.6203, 2.6248],
[ -0.6255, 0.1009, 0.7736, 0.4278]

            ],
            [
                [ -0.0000,   0.9916,  -0.1294,   1.2776],

[ -0.7802, -0.0809, -0.6203, 2.6248],
[ -0.6255, 0.1009, 0.7736, 0.4278]
],
[
[ -0.0000, 0.9916, -0.1294, 1.2776],
[ -0.7802, -0.0809, -0.6203, 2.6248],
[ -0.6255, 0.1009, 0.7736, 0.4278]

[kaihekaihe]

the depth video is ：

ego_Prior3.mp4

[KEH0T0]

This is expected behavior. Our training data does not enforce a perfect alignment between the Ego Prior and the Ego GT viewpoint — the model is trained to generate flexibly given an approximate prior, rather than reproducing the exact viewpoint.
That said, the generated view should still sufficiently reflect the direction of the prior you provided. Looking at your result, the output viewpoint appears reasonably aligned with the input prior, so the model is working as intended.