Error in Retraining pruned model

Accelerated Computing Intelligent Video Analytics TAO Toolkit
, , ,
1

Training a custom model on yolo for training is done and pruned also but after pruning while i retrain i am getting issue of
ValueError: Error when checking input: expected Input to have shape (3, 416, 736) but got array with shape (3, 544, 800)
terminate called without an active exception

• Hardware V100
• Network Type Yolo_v4
• TLT Version v3.21.11-tf1.15.5-py3
• Training spec file yolo_config.txt (2.3 KB)
• !tao yolo_v4 train --gpus 4
-e $SPECS_DIR/yolo_v4_train_cspdarknet53_kitti.txt
-r $USER_EXPERIMENT_DIR/exp
-k $KEY

2

Sorry, there is an known issue when retrain against pruned model with sequence format.

Please use tfrecord format instead. Thanks.

See more info in https://docs.nvidia.com/tao/tao-toolkit/text/object_detection/yolo_v4.html#tfrecords-format

4

@Morganh Can i use seq only for retraining and for training same tfrecords.

5

If train/retrain with a pruned model, it is expected to use tfrecord format.
If train/retrain without a pruned model, any format is ok.
Sorry for inconvenience. We already fix this issue and will be available for next release.

6

Sorry, I made a mistake. I modify my comment as above.

Similar topic: Training custom model using Yolo_v4_tiny - #15 by pierrenvz09

Currently, only tfrecord format can work when train/retrain with a pruned model.

7

@Morganh different error with this .

{{function_node __inference_Dataset_map__map_func_set_random_wrapper_7389}} /workspace/tao-experiments/yolov4_data/train/image/image/abc.jpg; No such file or directory

dataset_config {
data_sources: {
tfrecords_path: “/workspace/tao-experiments/yolov4_data/train/tfrecords/train*”
image_directory_path: “/workspace/tao-experiments/yolov4_data/train/image/”
}
include_difficult_in_training: true
image_extension: “jpg”
validation_data_sources: {
tfrecords_path: “/workspace/tao-experiments/yolov4_data/val/tfrecords/val*”
image_directory_path: “/workspace/tao-experiments/yolov4_data/val/image/”
}
}

8

I am afraid you need to modify

image_directory_path: “/workspace/tao-experiments/yolov4_data/train/image/”

to

image_directory_path: “/workspace/tao-experiments/yolov4_data/train”

9

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.