OPEN-AIR-SUN/mars

Key Error when training with KITTI-0002

Closed this issue · 5 comments

Hey, firstly thank you for making this great project public. I had some queries regarding training the model with KITTI-0002 dataset.
I'm getting a key error when I train it with KITTI-0002. I used both the latent codes : latent_codes_car_van_truck.pt and KITTI-MOT/car-object-latents/latent_codes02.pt but in vain. The error persists. This is also followed by a wandb 0.000mb of 0.013 mb completed error.

File "/usr/local/lib/python3.8/dist-packages/nerfstudio/engine/trainer.py", line 479, in train_iteration
_, loss_dict, metrics_dict = self.pipeline.get_train_loss_dict(step=step)
File "/usr/local/lib/python3.8/dist-packages/nerfstudio/utils/profiler.py", line 127, in inner
out = func(*args, **kwargs)
File "/app/mars/mars/mars_pipeline.py", line 135, in get_train_loss_dict
model_outputs = self.model(ray_bundle)
File "/usr/local/lib/python3.8/dist-packages/torch/nn/modules/module.py", line 1501, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.8/dist-packages/nerfstudio/models/base_model.py", line 142, in forward
return self.get_outputs(ray_bundle)
File "/app/mars/mars/models/scene_graph.py", line 445, in get_outputs
result = model.inference_without_render(ray_obj)
File "/app/mars/mars/models/car_nerf.py", line 206, in inference_without_render
self.car_latents[int(obj_id)].view(1, -1).to(obj_id.device),
KeyError: 0
wandb: \ 0.000 MB of 0.wandb: | 0.000 MB of 0.013 MB uploaded

I have made the changes in the cicai_config.py file and launch.json and set the depth to False for 0002. Can you suggest a fix for this? Works fine when training KITTI-0006 but fails for KITTI-0002

Hey, I'm running into the same issue. I'm trying to train on KITTI-0002, But it keeps giving me KeyError. @xBeho1der

Hi! @ng169 @amoghskanda

Thanks for your reply! This seems to be a known issue #16. If possible, we will train and upload a new latent code.

yeah sure, thank you!

The problem seems still exist in the KITTI-0002 0001, is latent code the updated version ?

Hi! @zwlvd

Thanks for your reply! The CarStudio paper is accepted by RAL and their code is available now at here. Maybe you can try randomizing a latent code or generate a latent code using Carstudio yourself. We greatly appreciate your help in fixing this issue!