How to produce cal file in export?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) RTX4080
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) Yolov4
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf1.15.5
I trained yolov4 in docker and tried to export onnx in int8 format using the following command.

yolo_v4 export -m /workspace/tao_tutorials/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18_epoch_080.hdf5 -k nvidia_tlt -e /workspace/tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4/specs/yolo_v4_train_resnet18_kitti.txt --batch_size 3 --data_type int8 --cal_image_dir /workspace/tao_tutorials/data/images --batches 10 --cal_cache_file /workspace/tao_tutorials/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18.cal --cal_data_file /workspace/tao_tutorials/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18.tensorfile

Only onnx and tensorfile are produced. There is no cal file. How to have cal file?

3

Do you mean the /workspace/tao_tutorials/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18.cal is not available?

4

Yes not produced in export. Only onnx and tensorfile are produced.

5

Could you double check? Please check with commands in tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4/yolo_v4.ipynb at main · NVIDIA/tao_tutorials · GitHub , access , etc.

6

Thanks.
I can produce onnx using the following command.

yolo_v4 export -m /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18_epoch_080.hdf5 \
                    -o /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights//yolov4_resnet18_epoch_80.onnx \
                    -e /workspace/tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4/specs/yolo_v4_train_resnet18_kitti.txt \
                    --target_opset 12 \
                    --gen_ds_config

But for the following command to produce int8 engine,

yolo_v4 gen_trt_engine -m /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights/yolov4_resnet18_epoch_80.onnx \
                                   -e /workspace/tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4/specs/yolo_v4_train_resnet18_kitti.txt \
                                   --cal_image_dir /workspace/tao_tutorials/StAndrew/data/images \
                                   --data_type int8 \
                                   --batch_size 16 \
                                   --min_batch_size 1 \
                                   --opt_batch_size 8 \
                                   --max_batch_size 16 \
                                   --batches 10 \
                                   --cal_cache_file /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights/cal.bin  \
                                   --cal_data_file /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights/cal.tensorfile \
                                   --engine_file /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights/trt.engine.int8 \
                                   --results_dir /workspace/tao_tutorials/StAndrew/data/experiment_dir_unpruned/resnet18/weights

I have error as yolo_v4 has no gen_trt_engine.

usage: yolo_v4 [-h] [--num_processes NUM_PROCESSES] [--gpus GPUS] [--gpu_index GPU_INDEX [GPU_INDEX ...]] [--use_amp] [--log_file LOG_FILE]
               {train,prune,kmeans,inference,export,evaluate,dataset_convert} ...
yolo_v4: error: argument /tasks: invalid choice: 'gen_trt_engine' (choose from 'train', 'prune', 'kmeans', 'inference', 'export', 'evaluate', 'dataset_convert')

I am using yolo_v4 from docker nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf1.15.5

7

Because you are using docker run way instead of tao_laucher. For tao launcher, you can find the docker info when you run $tao info --verbose. The gen_trt_engine is from tao-deploy docker, i.e., nvcr.io/nvidia/tao/tao-toolkit:5.5.0-deploy (TAO Toolkit | NVIDIA NGC). You can pull it.

8

ok i need to use deploy docker

9

Yes it works now