On 4.0.6 with a Zed2i I used a confidence threshold of 0 and got many valid depth values. On 4.0.7 I get only nan depth values. If I change the threshold to 1 I start getting valid depth values. Is this expected? If so please provide more details about how these new ranges should be interpreted and why it changed.
In a similar scene I get more valid values on 4.0.7 with confidence=1 than I do with 4.0.6 with confidence=0. If the depth model has improved great; however, If the model is giving more values that might be of lower quality that is problematic.
Neural. I understand the concept of the confidence values. I want to understand why performance and thresholds changed in 4.0.7 after remaining the same through 4.0.0-4.0.6.
Yes, I just received the confirmation that something changed for the NEURAL depth mode.
The confidence of the neural network was adjusted to more accurately estimate low-quality depth pixels, such as flying pixels.
We will add this missing information in the changelog of v4.0.7.