search

aharley / pips

Particle Video Revisited
MIT License
559 stars 49 forks source link

Feature representation not locked in test-time trajectory linking? #24

Closed ZHAOZHIHAO closed 1 year ago

ZHAOZHIHAO commented 1 year ago

Hi,

Section 3.7 of the paper says "To lock the model into tracking the “original” target, we simply re-use the original F0 across all re-initializations."

But in chain_demo.py, the query point's feature representation seems not locked?

line 54  outs = model(traj_e[:,cur_frame].reshape(1, -1, 2), rgb_seq, iters=6, feat_init=feat_init, return_feat=True)
line 57  feat_init = outs[3]

Best

HarryHsing commented 1 year ago

Hi, it will always be fixed on the initial ffeat, you may confuse it with ffeats

https://github.com/aharley/pips/blob/bde038a837c0ca94ad4c10748e57d7a3322565ac/nets/pips.py#L463

https://github.com/aharley/pips/blob/bde038a837c0ca94ad4c10748e57d7a3322565ac/nets/pips.py#L609

The return term in L609 is ffeat, not ffeats.

ZHAOZHIHAO commented 1 year ago

Thanks, you are right. I did confuse it with feats.