Xavi Diaz

Hi @CarkusL, can you give us the inference time for batch_size =1 of your TensorRT implementation including also the preprocess and postprocess?

Nevertheless, for me, it is still unclear how I can use centerpoint for **training** on my own custom dataset Can you guys provide more detailed steps, please? For training do...

Ok thanks a lot for your helpful indications! I have two questions more: 1. Since my training set is rather small, approx. 1200 lidar frames or less, I guess it...

Please correct me if I am wrong: a file like "infos_train_01sweeps_filter_zero_gt.pkl", has saved the aggregated ground truth information for all training frames at once, right? What's the meaning of the...

Hi, I have a couple more questions, if you could answer them please: 1. What is the sweeps attribute useful for? Why is it 10 on nuscenes and 1 or...

thanks a lot tianweiy as always for taking your time to reply, still have some questions: a) "For nuScenes, we use x,y,z,intensity and timestamp. For Waymo, we use x, y,...

Probably later on I would like to fuse camera images with the results from CenterPoint, following the state-of-the-art method in CLOCs. The available code released for now fuses the outputs...

I am aware that you are quite busy right now, but if you have some free time after ICCV, please try to answer a) of my previous comment. You don't...

Yes, I can show you in this gif how CenterPoint works after training on my own data recorded from the highway. The learned weights from nuscenes didn't help much, so...

Hello, I have one question about how the DataLoader works in the training script. I want to try Dynamic Voxelization, for which I need the points as input to the...