How to use the pretrained model

[Jaywxy]

I downloaded the pre-trained model you provided from Google, but I don’t seem to know how to use it. Can you help me?

I used the following command and it was prompted that the model file could not be found: python infer.py -d /autodl-tmp/Semantickitti/ -m ./model/LENet -l ./predictions -s valid and get error：

[Jaywxy]

I modified the code and now I get this error： This seems to be a missing file, but google drive only has the LENet file，It is not a folder, how do I need to unzip it?

[fengluodb]

see the https://github.com/fengluodb/RangeSeg/issues/5. I forget to zip config file with pretrained model. You can train the model from scratch or use the default config file to infer and train. I get the pretrained model by justing trainng serveral time.

[Jaywxy]

I read your tips in that issue. The above error occurred after I modified the two configuration file addresses in infer.py, so I don’t know how to modify the code.I just want to see the prediction time of your model on my machine. The problem now seems to be that the model cannot be obtained

[fengluodb]

train the model from scratch, then look the output dir's structure. You can move the pretrained model to the dir. Then, it may be run.

[Jaywxy]

Do you mean that the pre-trained model you gave cannot be used now? ? I need train a model myself?

[fengluodb]

when you run the train.py, you can get a dir. move the pretrained model to the dir. then, it's ok .

[Jaywxy]

I know ,okokok,thank you! Can you upload your trained model soon? thank you for your work！

[Jaywxy]

I tried to train a model yesterday, 2* 3090, batch_size=4, and then made predictions. I found that it did not match the infer time mentioned in your paper. I can only run to 13Hz, but your paper said that it can run to 25Hz. ? ? ? Is there something wrong with my settings? But I just changed the batch_size, and it was changed to a smaller size, and the batch_size will not be affected during infer.

[Jaywxy]

I tried to train a model yesterday, 2* 3090, batch_size=4, and then made predictions. I found that it did not match the infer time mentioned in your paper. I can only run to 13Hz, but your paper said that it can run to 25Hz. ? ? ? Is there something wrong with my settings? But I just changed the batch_size, and it was changed to a smaller size, and the batch_size will not be affected during infer.

And during training, it seems that only one core is working (NUM_workers is set). With the above configuration, it takes more than 4 days to complete 50 epochs. Can you give me some help? Thanks