search

bradyz / cross_view_transformers

Cross-view Transformers for real-time Map-view Semantic Segmentation (CVPR 2022 Oral)
MIT License
531 stars 81 forks source link

Pre-trained model #1

Closed Benzlxs closed 2 years ago

Benzlxs commented 2 years ago

Hi Bradyz, many thanks for sharing this amzing work, the idea is elegant. Currently, I am trying to use the code for 3D object detection task, but it takes long time to train the model. Would you mind providing well-pretrained model to speedup training process?

bradyz commented 2 years ago

Check out scripts/example.ipynb for a link to a pretrained model and how to load it

I'll update this in the main README soon along with some newer models

Is your code public? I can take a look Most models take about 8 hours on a 4 GPU machine

Benzlxs commented 2 years ago

Thanks.

The code is private reposity. Will let you know, if it is public in near future.

My input image size is 464*720, the batchsize is only 1 for each GPU, so it takes much longer time.

bradyz commented 2 years ago

Great, looking forward to it - I have not trained models at that resolution (the cross attention takes a lot of memory) so let me know what you do to get it to work