Testing the limits of a multi-task model with Intel Image Classification datasets, we split this dataset into two subsets (intel1 and intel 2), the first contains classes ('buildings', 'mountain', 'street') and the second contains ('forest', 'glacier', 'sea'), the idea is to build a single model for each task and compare it to the multitasking model with both tasks:
- configs/first_task.py : is the config for task 1 : intel 1
- configs/second_task.py : is the config for task 2 : intel 2
- configs/multi_task.py : is the config for multi tasking both task
- annotations for multi_tasking : download data and use script/construct_coco_multi_tasking.py
to train a model use the below command
python3 tools/train.py path_to_config --work-dir path_save_logs