ClearML with Nvidia TLT

This repository contains examples for using ClearML with the Nvidia TLT framework.

What's included?

• Train
• Evaluate
• Prune

Run TLT from the ClearML WebApp

See Run the examples. The example for NVidia TLT is located in the Nvidia TLT example project.

Manually running the examples

See Set up a ClearML account and Install ClearML Agent to set up your environment in case you don't have a locally-installed ClearML Server.

Train example

python train_tlt.py --module detectnet_v2 -m nvidia/tlt_pretrained_detectnet_v2:resnet18 \
        --arch detectnet_v2 --dataset-task <YOUR DATASET TASK ID> \
        --dataset-export-spec  example_specs/dataset_export_spec.txt -c specs/ \
         --key <Your key> --model_name model --experiment_spec_file  specs/detectnet_v2_spec_file_template.txt

Eval example

python evaluate_tlt.py --arch detectnet_v2 --dataset-export-spec example_specs/dataset_export_spec.txt \
                       --dataset-task <YOUR DATASET TASK ID> \
                       --experiment_spec_file example_specs/detectnet_v2_eval_kitti.txt \
                       --key <YOUR KEY> --train-task <YOUR TRAINING TASK ID>

Prune example

python prune_tlt.py --arch detectnet_v2 --trains-model-task <YOUR TRAIN TASK ID> --output_file \
                    /home/detectnet_v2/resnet18_nopool_bn_detectnet_v2_pruned.tlt --key <Your key>

When running each experiment, a new task will appear in the ClearML UI.