search

princeton-vl / px2graph

Training code for "Pixels to Graphs by Associative Embedding"
BSD 3-Clause "New" or "Revised" License
133 stars 22 forks source link

why should use munkres-tensorflow #3

Open hanghang2333 opened 6 years ago

hanghang2333 commented 6 years ago

hello,I am a beginner of this region.I dont know why should we use munkres-tensorflow. And,I can't download the pretrained model now,may be the superlink is dead.

anewell commented 6 years ago

Hi,

There's an operation in this code which requires doing a matching step with the Hungarian method. munkres-tensorflow provides an efficient implementation of that operation so it allows the code to run much faster.

The links for the pretrained models still seem to work for me?

ai-meng commented 5 years ago

hi, anewell, does this munkres-tensorflow calculate the gradient op? in my view, this so. file only calculates the Forward propagation result, but does not get the backward propagation result, so how did u use this package to optimize the loss?

thanks a lot!

meng

codezealot commented 5 years ago

hi, anewell, does this munkres-tensorflow calculate the gradient op? in my view, this so. file only calculates the Forward propagation result, but does not get the backward propagation result, so how did u use this package to optimize the loss?

thanks a lot!

meng

hi, ai-meng,

Did you find a way to calculate the gradient?