the meaning of graphs?

What are the specific meanings of the last four graphs generated?I don’t quite understand the meaning of the picture.

deeplabcut.plot_trajectories(config_path,[‘/analysis/project/videos/reachingvideo1.avi’])

generates several plots for an analyzed video. It creates a folder called ``plot-poses’’ (in the directory of the video). The plots display the coordinates of body parts vs. time, likelihoods vs time, the x- vs. y- coordinate of the body parts, as well as histograms of consecutive coordinate differences. These plots help the user to quickly assess the tracking performance for a video. Ideally, the likelihood stays high and the histogram of consecutive coordinate differences has values close to zero (i.e. no jumps in body part detections across frames).

(also, please don’t post the same question here and on GitHub. Here is the prefered questions platform!)

I could use some more info on the plots as well. For example - on one graph there are solid and dotted lines for each color – I think I’ve surmised that one is the X and one is the Y coordinate – but why am I surmising? Is there a doc somewhere? A legend on the plot would be very handy too if its not hard to do.

good point… I will add to the next release :wink: @JohnJCraig @liuxin

The plots display the coordinates of body parts vs. time, likelihoods vs time, the x- vs. y- coordinate of the body parts, as well as histograms of consecutive coordinate differences. These plots help the user to quickly assess the tracking performance for a video. Ideally, the likelihood stays high and the histogram of consecutive coordinate differences has values close to zero (i.e. no jumps in body part detections across frames).

Could you explain the meaning of ‘Consecutive coordinate difference’ graph in more detail? ‘Everybody part likelihood over time’ graph shows which model achieved low performance. However, I can’t feel the difference between high performance model and low performance model in the ‘Consecutive coordinate difference’ graph.

hist hist (4)
I’m sorry if you don’t get the meaning of what I said. I am a foreigner who is not good at English.

@JE_J As the axis label says it is delta (change) of X and Y. Shows you what was the pixel distance between labels in consecutive frames. Like, you can see that most are like <10 px, which is around what you’d expect from a moving animal, but you can also see some less occuring high deltas (>20 px), which means some of your frames have jumps between them that probably cannot happen naturally hence, they stem from model giving bad detection in those frames. You can see that the top plots (for both likelihood and coordinate difference) are showing model that’s better - it has higher overall likelihood, and lower overall deltas. In other words, it knows better what to label as your bodypart. Hope this explanation clears it out

1 Like