requirements

Requirements

Detectron 2 Training

Install relevant Detectron 2 version after checking cuda and torch versions.

To reproduce the enviroment, there are several options provided to try (arranged as clean to bloated):

req_pigar.txt

• Uses Pigar package to generate only the requirements in accordance with imports in code.
• Only considers the imports of the specific repo (Detectron2_train in this case)
• This boils it down to the minimum installations and reduces possibilities of conflicts.
• The txt file can be used normally via pip to recreate environment
• You may require to approve and/or initiate further installations if prompted, depending on your base installations.

req_pipreqs.txt

• Uses pipreqs package to generate requirements from imports.
• --use-local flag was used.

env_no_builds.yml

• conda export without build strings
• suggested here
• Useful for OS-agnostic installation

More bloated installations : (irrelevant packages could be present)

env.yml

• usual full conda export of environment with builds specific to original platform.

req_piplist.txt

• frozen pip list.
• suggested here

req_pipfreeze.txt

• usual pip freeze of full environment, could fail to resolve directories.
• only records and saves the packages that were installed with pip install.