- I am currently working on a data set which has the structure - frames x Number of Objects.
- Each object is of size 7x5 and the task is to classify each object into one of the 4 classes.
- I have been getting good results using CNN but we also have a temporal component in our data and hence want to try out convLSTM.
- The Problem - Each frame in the data set has a timestamp and is temporally in sequence but the objects within a frame are not related in the time domain. How could I restructure my data so that I could use convLSTM/LSTM and predict class for each object? Basically I want to explore temporal relationship between predictions of two consecutive frames instead of looking at predictions of two consecutive objects.
- I would appreciate any help/suggestion/tip.
$\begingroup$
$\endgroup$
