Feng Wang
Feng Wang
Waymo dataset is annotated on 200 frames of point clouds at the same time. Could this tool illustrate and annotate over 10 million points in one scene?
As far as I know, the FN is actually Batch Normalization applied after final feature layer. I used to do such experiments about one year ago because we usually do...
Some users reported that the program would crash when using triplet loss. Similar issue https://github.com/BVLC/caffe/issues/2995 . It seems that the caffe_gpu_dot() is unproperly used. I have merged another implementation of...
A simplest reproducible code is here: ``` import numpy as np import torch from mmdet3d.core.bbox.structures import LiDARInstance3DBoxes boxa = np.array([[ 2.38607935e+02, 9.13623668e+02, 8.59954819e-01, 1.85358202e+00, 4.57165861e+00, 1.63146234e+00, 3.02226729e+00]]) boxa = torch.tensor(boxa).view((1,-1))...
Thanks for your inspiring paper. I generated the rpn features using the provided pth file and followed the default configs (ROI_SAMPLE_JIT: True). The rcnn evaluation results are: ``` Car [email protected],...
In formulation (3), there is a factor \gamma. This parameter is setted to prevent the hinge loss to be 0. However, I haven't find this parameter in the code. The...
I noticed that the weight of the loss layers are realised by setting the blob_lr parameter of the innerproduct layer in this project. This is equivalent to the formulation (3)...
When I fp8 quantize a model and then shard it using FSDP2, it reports an error: ``` [rank1]: Traceback (most recent call last): [rank1]: File "/mnt/teams/algo-teams/shared/code/wanx-inference/generate.py", line 461, in [rank1]:...