search

TRI-ML / PF-Track

Implementation of PF-Track
Other
208 stars 27 forks source link

Question about motion prediction #8

Closed jianingwangind closed 1 year ago

jianingwangind commented 1 year ago

The code structure is really elegant!

And i have a question w.r.t. motion prediction/ref points update:

the reference points is updated twice in tracker.py with motion predictions and ego movements respectively, meaning the predicted motion is defined in the current frame coordinate system?

ziqipang commented 1 year ago

@jianingwangind Yes. I remembered that the predictions on every timestamp is the single-frame movement $(\Delta x, \Delta y)$ located in the local sensor coordinate.

jianingwangind commented 1 year ago

Thanks a lot.

jianingwangind commented 1 year ago

Another question maybe...

In history refine, ref points and bboxes are updated and they are used to update the corresponding cache_xxx, but why not update the ones in history bank? I saw you just comment them.

ziqipang commented 1 year ago

@jianingwangind I evaluated the code before releasing the code, but I didn't remember and notice what you mentioned, to be honest. I guess the current implementation considers that a bounding box with a low score shouldn't fit into history. Thus, the predictions will not be hist_xxx directly.

However, I think that your suggestion is also reasonable. As I don't have enough computation right now, do you mind uncommenting Line 135 and 138 of spatial_temporal_reasoner.py and see how things will go?

jianingwangind commented 1 year ago

Sure, thanks for your reply:)