Devlopement

implement building action tubes in python given frame label detection based on ICCV17 paper
fill the gaps in genrated tubes;
Write evaluation protocal for tube of all types;
fix detection scaling, i.e. normlise the output of network to [0-1]
write frameAP evaluation All 6
[] Generate tubes with marginals for duplexes and triplets
add clssifcation for ego action classification for AV-actions
frame-ap evaluation for AV-actions

Fix annotations

integrate last video from reza, video annotate by alex bruce but correction not complete.
Redefine splits - remove the video the shorter video from evaluation
integrate last video from Andy
Fix start and end of the interploated videos @12fps from old version of VOTT

Experimentst

[] run experiment with different input sizes
run experiment with different network achtecture
[] see if we can accomodate RCN style YOLOv5
[] check if giving information about anchor-location help in determining location better

Major devlopments

[x] accomdate multiple frames as inputs