Open3DSOT
Open3DSOT copied to clipboard
Ghostish
Excuse me, about the reproducing of Pedstrian and Van on Kitti.
Dear authors,
Great work and paper! I am a postgraduate student and my research project is 3D SOT. Regarding the result of reproducing on the KITTI dataset, I found that the results I reproduced on the Car and Cyclist data sets were basically the same, but the results on Pedestrian and Van were very different.
The reproducing success/precision on Pedestrians are near 50/75, much lower than the results in the paper 60/89.
I just change the dataset name in the yaml cfg file (Car to Pedestrian) and then train it. I would be appreciated if you can share me more details on the model parameter setting if i fogot to set.
Thank you very much!
Can you reproduce the performance when using the provided checkpoint in this repo? If not, there may be something wrong with your data.
Can you reproduce the performance when using the provided checkpoint in this repo? If not, there may be something wrong with your data.
Yes, i can reproduce the performance with the pretrained model, but if i train the model ( change the dataset name in the yaml cfg file, from car to pedestrian), my test result will be low ┭┮﹏┭┮
Can you reproduce the performance when using the provided checkpoint in this repo? If not, there may be something wrong with your data.
So i just want to know if i have to change anything else except the category_name:
Can you reproduce the performance when using the provided checkpoint in this repo? If not, there may be something wrong with your data.
So i just want to know if i have to change anything else except the category_name:
And the car & cyclist 's results will be similar to the results of the paper, but Pedestrian and Van are much lower ┭┮﹏┭┮
Thanks very much, i have already seen #33 and i found the preload data was 1.3GB. i will try to type a smaller bs when start the training.