Hi, thank you for the great work and for sharing the code for Align3R.
I am a researcher studying your paper and have a question regarding the implementation details.
In Section 4.1, you mentioned that you used s = 6 layers of features in Equation (3).
Was the value of s=6 chosen based on a specific ablation study comparing different numbers of layers (e.g., s=4 or s=8)?
If so, did you find s=6 to be the optimal balance between performance (depth/pose accuracy) and computational efficiency?
In your experiments, did you observe any significant performance changes when varying the number of injection layers?
Thank you for your time and insights!
Hi, thank you for the great work and for sharing the code for Align3R.
I am a researcher studying your paper and have a question regarding the implementation details.
In Section 4.1, you mentioned that you used s = 6 layers of features in Equation (3).
Was the value of s=6 chosen based on a specific ablation study comparing different numbers of layers (e.g., s=4 or s=8)?
If so, did you find s=6 to be the optimal balance between performance (depth/pose accuracy) and computational efficiency?
In your experiments, did you observe any significant performance changes when varying the number of injection layers?
Thank you for your time and insights!