Running on custom data

[Ritchizh]

Hi! I’m trying to run tester_SemanticKITTI with weights from pretrained model on my custom data. I was able to do it for tester_S3DIS and custom indoor data, and wanted to test some outdoor data with no RGB values now. I edited scripts to keep only ‘test’ part, for example in init_input_pipeline() I keep: gen_function_test, gen_types, gen_shapes = self.get_batch_gen('test') instead of self.get_batch_gen('training'). When running tester.test(model, dataset), I get:

Model restored from pretrained_models/SemanticKITTI/snap-277357
Epoch 0, end. Min possibility = 0.0

And it keeps looping never meeting condition of np.min(dataset.min_possibility) > 0.5. Could you please give any clues how to test what is wrong with model/data initialization?

[Marcelhaczyk]

Hello, did you solve this issue ? I'm facing the same problem

[ZheningHuang]

Same problem here. Did you guys find the solution

[ZheningHuang]

Found the problem, make sure you change the val_batch_size in the helper_tool.py

[Ritchizh]

Found the problem, make sure you change the val_batch_size in the helper_tool.py

Hi! val_batch_size = 20 Do you set it to another value?

[ZheningHuang]

Try to change it to 1, if you have only 1 point cloud to test. I remember there is a calculation that used this variable and if you do not have more than 20 (val_batch_size) point clouds to test it will cause a problem.

[Ritchizh]

Try to change it to 1, if you have only 1 point cloud to test. I remember there is a calculation that used this variable and if you do not have more than 20 (val_batch_size) point clouds to test it will cause a problem.

Thank you for reply. I tried val_batch_size=1, it did not work for me.

[hansgrammer]

could you tell me how to use the data with no RGB? Convert colors to uniform values？my data only have xyz and intensity with no RGB .i don't no how to use this data.

[Jeehoon-K]

could you tell me how to use the data with no RGB? Convert colors to uniform values？my data only have xyz and intensity with no RGB .i don't no how to use this data.

Hi, you can just add RGB values as (0 0 0) on your data

[hansgrammer]

thank you and that's what I did

At 2022-03-24 22:22:56, "Jeehoon, Kim" @.***> wrote:

could you tell me how to use the data with no RGB? Convert colors to uniform values？my data only have xyz and intensity with no RGB .i don't no how to use this data.

Hi, you can just add RGB values as (0 0 0) on your data

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[zzlqwq]

So has anyone solved this problem?

[hansgrammer]

which problem？

在 2022-12-09 20:57:28，"张家乐" @.***> 写道：

So has anyone solved this problem?

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[zzlqwq]

I'm running the Semantic3D test provided by the author. I meet the same problem as the building owner. The output of the terminal is as follows:

Model restored from /home/zjl/Desktop/models-20221208T125738Z-001/models/Semantic3D/snap-37001 Epoch 0, step 1. min possibility = 0.0 Epoch 0, step 2. min possibility = 0.0 Epoch 0, step 3. min possibility = 0.0 Epoch 0, step 4. min possibility = 0.0 Epoch 0, step 5. min possibility = 0.0 Epoch 0, step 6. min possibility = 0.0 Epoch 0, step 7. min possibility = 0.0 Epoch 0, step 8. min possibility = 0.0 Epoch 0, step 9. min possibility = 0.0 Epoch 0, step 10. min possibility = 0.0

I don't understand why this is happening.