Missing arch_cfg, data_cfg in Pretrained Weights.

[AbhishekKaushikCV]

Great Work!!,

• From where, I can download the pretrained weights? Is it public yet?

[fengluodb]

Thank you for your attention for this project. The arch_cfg and data_cfg are just in config/arch and config/labels.

I will also update the download link of pretrained weights, adding the arch_cfg and data_cfg.

[AbhishekKaushikCV]

Thank you for the quick response.!!

I have few more questions regarding the training and inference on the custom dataset (Dataset).

I converted my dataset into the .bin files and followed the structure of the SemanticKITTI.

• What is the significance of calib.txt, and poses.txt in the training and inference of LENet? I don't have these files for my custom dataset. Do I need the calib.txt, and poses.txt for inference and training on my custom dataset?
• What is the the effect of the intensity feature (the fourth column in the semantickitti) on the performance of the model? In my dataset, the intensity range is not in [0, 1]. Do I also have to normalize the intensity in my custom dataset?
• What is the efficient way (changes) to update the LENet code for the inference and training on my custom dataset?

[fengluodb]

Thank you for the quick response.!!

I have few more questions regarding the training and inference on the custom dataset (Dataset).

I converted my dataset into the .bin files and followed the structure of the SemanticKITTI.

• What is the significance of calib.txt, and poses.txt in the training and inference of LENet? I don't have these files for my custom dataset. Do I need the calib.txt, and poses.txt for inference and training on my custom dataset?
• What is the the effect of the intensity feature (the fourth column in the semantickitti) on the performance of the model? In my dataset, the intensity range is not in [0, 1]. Do I also have to normalize the intensity in my custom dataset?
• What is the efficient way (changes) to update the LENet code for the inference and training on my custom dataset?

1. the calib.txt and poses.txt don't influence the inference and training on custom dataset. the calib.txt and poses.txt are used for the labeling tool.
2. As for the intensity feature, I think it's better to normalize it.
3. As for the inference and training on your custom dataset, if your dataset format like SemanticKITTI, you can used the way I train and infer on SemanticKITTI. And You need to write a data_cfg like SemanticKITTI and update the code https://github.com/fengluodb/LENet/blob/0b9f62032a0a53141e6fc925ae410473e94d795a/config/arch/LENet.yaml#L68-L88 I think that should be ok.