I'm using the model to extract camera trajectory data for integration with Unreal Engine. However, I'm encountering an issue where the output data appears to be nearly zero or does not produce a meaningful camera motion path.

Details:

Data Sample:
For example, processing a test video (limited to 17 frames due to memory constraints) produced the following output:

python-repl
Copy
Edit
0.0 0.0001748965005390346 -2.4827617380651645e-05 -0.003876725910231471 0.9999041012894861 0.009460109122497035 0.010066458405044089 0.0009802933412098088
1.0 0.00019531168800313026 -0.0001670413330430165 -0.002075210213661194 0.9998958531657915 0.009938693622128361 0.010415446426130083 0.0010117641443218369
...
16.0 -0.003995809238404036 0.0016624971758574247 0.02769709751009941 0.9993528576808245 0.02428708886281234 0.02630896892058428 0.0034411210492250517
The first column is the frame/time stamp. The subsequent columns represent the camera’s translation and rotation data (likely in quaternion form).

Issue Observed:
When running multiple videos, the extracted camera trajectory data remains nearly zero (or with extremely small values) which fails to produce effective camera tracking in Unreal Engine.

Questions:

Is this near-zero output expected given the limited number of frames, or is there a potential issue with the model's output?
Am I misinterpreting the data format (i.e., the meaning of each column, especially the rotation data)?
What would be the best approach to extract a meaningful and accurate camera trajectory, especially when constrained by memory limitations?
The output trajectory data is in which units? meters or centimeter?
Any clarification or recommendations to resolve this issue would be greatly appreciated.