search

shedy-pub / hlagcn-jittor

Jittor implementation of the paper "Hierarchical Layout-Aware Graph Convolutional Network for Unified Aesthetics Assessment"
50 stars 7 forks source link

Pretrained Weights #2

Open koustav123 opened 2 years ago

koustav123 commented 2 years ago

Hi,

Are there any pre-trained weights available for AADB dataset?

I tried to run the training on AADB dataset. I get this warning both during training and testing. And my numbers are 0.66/0.63 on AADB, much lower than reported in the paper (0.90/0.89). I did not change anything else in the code.

Also, I am running the code on a single GPU in a virtual environment on a single PC without mpi.

image

Any pointers will be highly helpful.

Thanks

evenhax commented 2 years ago

Got the same problem, looking forward to your reply.

islandLZ commented 2 years ago

Hello, excuse me, MSE, SRCC, LCC, ACC, EMD_ 1 、EMD_2 ，What do these represent respectively? And how to calculate the score?

flyingbird93 commented 2 years ago

Have you ever encountered a situation where the loss output is always "nan"? If so how did you solve it?

Do you ensure proper use of train, validation and test sets during training? Probably this is why the results are poor on AADB.