Train yolov4 with custom dataset

[pardis]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU) Jetson Orin Nano
• DeepStream Version 6.3
• JetPack Version (valid for Jetson only) 5.1.2
• TensorRT Version
• NVIDIA GPU Driver Version (valid for GPU only)
• Issue Type( questions, new requirements, bugs)

I am training Nvidia models on the dGPU with nvidia tlt stream analytics toolkit docker version 3.21.
I want to train a yolov4 with custom dataset of different shapes. but it gives me this error:

ValueError: Negative dimension size caused by subtracting 7 from 6 for 'conv1/convolution' (op: 'Conv2D') with input shapes: [?,3,6,6], [7,7,3,64].

I tried to convert my dataset to tfrecord and it gives me this error

Message type "DataSource" has no field named "tfrecords_path".
Traceback (most recent call last):

can you please help me with that
Thanks

[Morganh]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks

@pardis

tlt stream analytics toolkit docker version 3.21.

For YOLOv4, could you please use the latest tao-tf1 docker for it?
nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf1.15.5 from https://catalog.ngc.nvidia.com/orgs/nvidia/teams/tao/containers/tao-toolkit/tags.

And its notebook and spec file can be found in tao_tutorials/notebooks/tao_launcher_starter_kit/yolo_v4 at main · NVIDIA/tao_tutorials · GitHub.

Source code can be found in tao_tensorflow1_backend/nvidia_tao_tf1/cv/yolo_v4 at 2ec95cbbe0d74d6a180ea6e989f64d2d97d97712 · NVIDIA/tao_tensorflow1_backend · GitHub.