volvo-cars / fb-ssem-dataset Goto Github PK

hope the author can share some code of result visualization ...thx

LIke this in the dataset ( xxxx.txt ):

BEVCamera, 7.467841, 40.13262, 2.768845, 90, 0.9908196, 0

thanks for sharing this project, i also have real world fisheye images to predict but got no GT, may i ask when will you release the style transfer tools and process?

as mentioned in the paper,

We run neural style transfer [41] on the simulated FB-SSEM dataset images using samples of realworld images as style images. Additionally, we note that the simulated fisheye cameras differ from those mounted on the actual car in terms of the valid regions of the image (i.e., regions not covered by the camera or the ego car). We account for this difference during training by predefining the valid region for every camera in the training data to be the same as that of the corresponding real-world camera.

after training 1 epoch, the process didn't move on...

Train Epoch: 0 [7007/7010 Srno: 1 (100%)] Loss: 0.207433
Train Epoch: 0 [7007/7010 Srno: 2 (100%)] Loss: 0.154045
Train Epoch: 0 [7008/7010 Srno: 0 (100%)] Loss: 0.298333
Train Epoch: 0 [7008/7010 Srno: 1 (100%)] Loss: 0.299832
Train Epoch: 0 [7008/7010 Srno: 2 (100%)] Loss: 0.261416
Train Epoch: 0 [7009/7010 Srno: 0 (100%)] Loss: 0.131326
Train Epoch: 0 [7009/7010 Srno: 1 (100%)] Loss: 0.203777
Train Epoch: 0 [7009/7010 Srno: 2 (100%)] Loss: 0.171052
Train Epoch: 0 [7010/7010 Srno: 0 (100%)] Loss: 0.096523
Train Epoch: 0 [7010/7010 Srno: 1 (100%)] Loss: 0.097764
Train Epoch: 0 [7010/7010 Srno: 2 (100%)] Loss: 0.110357
Train Error: Avg loss: 0.232530

Test Error: Avg loss: 0.062938

The code of F2BEV will release?

This is great work, could you please provide the trained model weights? My email: [email protected], thank you very much!

Looking forward to using the model weights mentioned in the paper for indicator validation and inference experiments， thank you.

I would like to use FB-SSEM dataset for a task that requires BEV rgb images but these images are missing for almost all sequences (except Images0 and Images1). Will the remaining BEV images be released ?

Thank you in advance for your answer and sharing this amazing work !