adding images

final commit

latest pretrained model changes half and notebook

reduced batch_sizes(mem error due to inc history) by half and doubled val_len, plot_freq and smoothing

reverted history to 10/0.1, increased plot smoothing freq to 2000, training from scratch

smoothing function for training plot, 1000 steps for val, plot every 3000

adding line to import perception labels

changed raster size to rectangle

added the changes to the first training loop validation portion!

adding disbale traffic lights flag and implemented neg log likelihood loss func

removing the transform from the validation code

removing transform and training on original targets

adding transfor code

not sure what improvements

pin_memory, no_grad, plotting colour and batch_size