VizWiz VQA course project Multi Modal Machine Learning

Running Instructions

1. Download data:

Download skill data:

cd data/skill
bash download_data.sh

Download VQA data:

cd data/VQA
bash download_data.sh

2. Run model (SkillCLIP) variants:

With everything:

python -m src.main_model.clip_late_fusion -t -de "cuda:0" -exp skill_aware_clip

Without skill embeddings:

python -m src.main_model.clip_late_fusion -t -de "cuda:0" -exp skill_unaware_clip

Without object tags:

python -m src.main_model.clip_late_fusion -t -de "cuda:0" -exp skill_aware_clip_nobj -nobj

Without scene text:

python -m src.main_model.clip_late_fusion -t -de "cuda:0" -exp skill_aware_clip_nsctxt -nsctxt

With multi-task training:

python -m src.main_model.clip_multitasking.py -t -de "cuda:0" -exp skill_aware_clip_multitasking -pred_file pred.json

Interesting object detections

Keys of a keyboard are detected as microwaves with relatively high confidence scores:

1. path: val_objects_detected/VizWiz_val_00001474_objects.png
   Potential reasons: the image is very zoomed in which might be abnormal.

Illustrative Examples:

Here are some illustrative examples from our error analysis: FusionCLIP refers to the SkillCLIP model without the skill embeddings. Comparison between our model (SkillCLIP) and FusionCLIP. Some more examples: