Semantic Analysis of Cultural Heritage Data:Aligning Paintings and Descriptions inArt-Historic Collections

Code for our paper [Semantic Analysis of Cultural Heritage Data:Aligning Paintings and Descriptions inArt-Historic Collections](TODO add link)(Link will be added) that will be published at the "International Workshop on Fine Art Pattern Extraction and Recognition (FAPER)" held in conjunction with ICPR 2020.

This README is only a stub for now. More information about how to get the datasets and how to train a model, are coming soon!

Start training

• CUDA_VISIBLE_DEVICES=0,1 python -m torch.distributed.launch --nproc_per_node=2 main.py
  • Change nproc_per_node to number of GPUs used
  • Change CUDA_VISIBLE_DEVICES to number of GPUs used
  • If a training is already running specify a different port with --master_port=$RANDOM

Parameters

• --coco path to CoCo root
• --semart path to SemArt root
• --wpi path to WPI evaluation data root
• --resnet either 'pretrained' or path to ResNet152 checkpoint file
• --batch-size
• --epochs
• --learning-rate
• --mmd-weight factor to multiply MMD loss with (default: 1)
• --log-dir TensorBoard log dir
• --reduction 'sum' or 'mean' (how to reduce the supervised/Hinge loss)
• --load-model path to checkpoint file