Lesson 3 Regression Application - Object Localization

After working through the Head Pose data as taught by Jeremy in Lesson 3, I tried implementing the regression concept for object localization problem using a custom dataset of everyday objects and to find a bounding box for the same. The problem required the model to predict 4 continuous values that are basically the coordinates of the vertices on the primary diagonal of the bounding box.But while creating the Databunch using the data block API I ran into this error:

As per my understanding, I think this is due to the change in the number of regression points that are to be predicted as in the Head pose data the label tensor had two values and in this case, it has 4 values. So I thought I just have to change the shape of the tensors in some function. But I don’t know where to change and what to change. Can anyone help me figure it out or is there any other way to approach this problem

1 Like