From left to right (`left_image`, `right_image`, `valid_mask`, `valid_mask & ground_truth`, `prediction`). **Darker is further away, lighter is closer**. In the case of `Sintel` which is more closely aligned to the original distribution of `CREStereo` we notice that the model accurately predicts the background scale whereas in the case of `Middlebury2014` it cannot correcly estimate the continous disparity. Notice that the frame composition is similar for both examples. The blue skybox in the `Sintel` scene behaves similarly to the `Middlebury` black background. However, because the `Middlebury` samples comes from an extremly large scene the crop size of `384x512` does not correctly capture the general training distribution.
...
...
@@ -155,7 +155,7 @@ From left to right (`left_image`, `right_image`, `valid_mask`, `valid_mask & gro
The top row contains a scene from `Sceneflow` using the `Monkaa` split whilst the bottom row is a scene from `Middlebury`. This sample exhibits the same issues when it comes to **background estimation**. Given the exagerated size of the `Middlebury` samples the model **colapses the smooth background** of the sample to what it considers to be a mean background disparity value.
For more detail on why this behaviour occurs based on the training distribution proportions you can read more about the network at: https://github.com/pytorch/vision/pull/6629#discussion_r978160493
...
...
@@ -173,7 +173,7 @@ In **grey** we set the lr decay to begin after `30000` steps whilst in **orange*
##### Gradient norm saves time
In **grey** we keep ``gradient norm`` enabled whilst in **orange** we do not. We can notice that remvoing the gradient norm exacerbates the performance decrease in the early stages whilst also showcasing an almost complete collapse around the `60000` steps mark where we started decaying the lr for **orange**.
