Apply PREVALENT pretrained code on HANNA task. Clearing the redundant code and unused definitions. You can still try the code now.
Environment setting up: 1 sudo bash scripts/build_docker.sh Possible issues: Step 1/17 : FROM nvidia/cudagl:10.0-devel-ubuntu18.04 Get https://registry-1.docker.io/v2/: dial tcp: lookup registry-1.docker.io on [::1]:53: read udp [::1]:39620->[::1]:53: read: connection refused
You need to log in your dockerhub account to pull images
2 set your environment 2.1 Export MATTERPORT_DATA_DIR=/home/victor/work/vlnwork/Matterport3DSimulator/skybox_images/v1/scans
2.2 create your container sudo -E bash scripts/test_run_docker.sh
2.3 build simulator bash scripts/build_simulator.sh
Install missing dependencies: apt-get install python3-venv python3-pip pip3 install pytorch_transformers==1.2.0
2.4 test if simulator is ready python scripts/test_api.py
2.5 modify define_vars.sh # set your folder to store the result models modify train_main.sh # set your gpu id and output folder name modify config/hanna.json # training parameter setting cd tasks/HANNA/exp_scripts bash train_main.sh
2.6 to test your saved model bash eval_main.sh unseen_all