VaianiLorenzo / VWSD

This repository contains the code to setup the experiments for the SemEval 2023 Visual-Word Sense Disambiguation task.

PoliTo Team @ SemEval 2023 Task 1: Visual-Word Sense Disambiguation

This repository contains the code to setup the experiments for the SemEval 2023 Visual-Word Sense Disambiguation task.

Finetuning

To finetune the the standard CLIP large pretraing from Hugging Face (here), you can run:

$ python clip_finetuning.py \
  --textual_input full_phrase \
  --log_filename clip_training.txt \
  --epochs 30 \
  --batch_size 16 \
  --model_size large \
  --textual_augmentation \
  --visual_augmentation

Inference

If you want to test the standard pretrained CLIP model from Hugging Face (both large and base) you can run:

$ python single_clip_inference.py \
  --log_filename clip_results.txt \
  --log_step 200 \
  --phase val \
  --model_size large

Otherwise, if you alredy finetuned a model, you can load and use your checkpoint, stored in the checkpoints folder, running the following:

$ python single_clip_inference.py \
  --log_filename clip_results.txt \
  --log_step 200 \
  --phase val \
  --clip_finetuned_model_name clip_finetuned.model