What type of image conversion does nvds_obj_enc_process perform?

[hyperlight]

Please provide complete information as applicable to your setup.

• Hardware Platform (Jetson / GPU): GeForce RTX 3090
• DeepStream Version: 5.1
• TensorRT Version: 7.2
• NVIDIA GPU Driver Version (valid for GPU only): 470
• Issue Type( questions, new requirements, bugs): questions

I can’t find the source code for nvds_obj_encode.h, only the shared library associate with it. Is the source code for nvds_obj_encode.h open sourced?

I use nvds_obj_enc_process to crop detected objects and save them as jpeg images for further training of the models I use in my pipeline (I use the tao toolkits). I wonder if any internal image conversion done by nvds_obj_enc_process need to be taken into consideration during training and/or inference?

[fanzh]

1 nvds_obj_enc_process is not opensource.
2 nvds_obj_enc_process will encode raw data to jpeg, please refer to nvds_obj_encode.h, after decoding these data can be used to train and inference again.

[hyperlight]

Hi @fanzh,

Thank you for the response, the nvds_obj_encode.h header file just mention “JPEG encode”, by “decoding” do you mean the jpeg files saved need to be pre-processing in some way to before they can be used for re-training in tao or can I just use the jpeg files directly?

[fanzh]

model only accept raw data, jpeg is encoded data, so you need decode jpeg file to raw data, then feed raw data to model.
If there are any TAO questions, I suggest opening a new topic in the TAO forum.