quancore / social-lstm

Social LSTM implementation in PyTorch

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Inference on plots from vizualize.py -Tobs & Tpred unknown

avijitashe opened this issue · comments

I'm trying to understand what the plots depict? There are 20 frames, at interval of 10, at the top with one pedestrian plotted with "target ped 24 pred." and "target ped 24 true" which should mean the ground truth vs predicted traj. (THE SAME FOR ped 111 and so on... in every plot)

But, from which frame/time it starts to predict? That is, what is t=1 to Tobs, and Tobs+1 to Tpred? The paper says, it observes for 8 frames and predicts for next 12. So, which are these 8 and 12?

Kindly help. Thank you!

Here are the plots

sequence00003
sequence00028

The information you gave at the beginning is correct. Basically frame_ids list top of all plots indicates which frame-group we have focused on (listing all frames in a frame-group). Each frame-group includes 20 frames (8+12) and the plots are illustrating the target ped's true and predicted trajectories with its neighbors existed in this frame-group (the number of vertex in a neighbor trajectory can vary because not all neighbors existed in all frames). The vertexes on a trajectory have not been numbered so you cannot know which part is observed and which part is predicted (Feel free to create a pull request enumerating the points in a trajectory by using frames). The main purpose of this plot is how related the ground truth and predicted trajectories overall, not frame by frame.

How can I do that (Numerating the points in trajectory as observed and predicted)? Where to look for the required files? Is this data being stored in any CSV somewhere?

Regarding this numerating of "observed" and "predicted": As I understand the prediction part, it happens like this, kindly correct me if I'm wrong - 1) The LSTM observes for the first 8 frames, does not make any prediction before it has observed 8 frames 2) On receiving 8, on 9th frame, it predicts for next 12

Hence, in each of the plots, the vertices for "true" should be more than "pred". Because "pred" starts after 8 frames, the vertices for "true" should appear first for some frames, and when 8 frames have been received, the vertices of "pred" and "true" should appear together, where we see how mush is the error