Many thanks for releasing your implementation and dataset. I have found that there seems to be discrepancy between the range of camera poses described in the paper and those in the released dataset.
In section 3.3 Data Collection, under the heading "Construction of the Multi-View Video Data" it states "we constrain the elevation of each camera between 0◦ - 45◦, and the azimuth between 0◦ - 360◦."
However in the released SynCamDataset I have found that camera azimuths range between +-30◦, and elevation ranges from 0 to 30◦.
This means that we cannot train the model to generate synchronised videos with larger angles such as a 90◦ Difference in Azimuth (as shown in paper figure 5) or 120◦ Difference in Azimuth (Figure 7) or 45◦ Difference in Elevation (Figure 1). Would you be able to release the full dataset, so that other researchers can try to reproduce those results?
Many thanks for releasing your implementation and dataset. I have found that there seems to be discrepancy between the range of camera poses described in the paper and those in the released dataset.
In section 3.3 Data Collection, under the heading "Construction of the Multi-View Video Data" it states "we constrain the elevation of each camera between 0◦ - 45◦, and the azimuth between 0◦ - 360◦."
However in the released SynCamDataset I have found that camera azimuths range between +-30◦, and elevation ranges from 0 to 30◦.
This means that we cannot train the model to generate synchronised videos with larger angles such as a 90◦ Difference in Azimuth (as shown in paper figure 5) or 120◦ Difference in Azimuth (Figure 7) or 45◦ Difference in Elevation (Figure 1). Would you be able to release the full dataset, so that other researchers can try to reproduce those results?