The process of learning this wonderful project 'OpenCOOD' includes studying the code and comments in the source files.
Rujia Wang email: [email protected]
Jinhao Xu email: [email protected]
branch main(Rujia Wang):
2023-06-05 train the point_pillar_v2vnet batchsize=2 original but got the all zero results
2023-06-04 train the point_pillar_intermediate_fusion batchsize=6
2023-06-03 train the point_pillar_late_fusion batchsize=20
2023-06-02 train the point_pillar_fcooper batchsize=6
2023-06-02 train the point_pillar_early_fusion batchsize=16
2023-05-13 opencood/models/sub_modules/pillar_vfe.py
pillar voxel feature encoding
inputs: voxel_features(体素特征 [n,32,4]) voxel_num_points(每个体素之内包含几个点 [n]) coords(每个提速的坐标 [n,4])
outputs: batch_dict['pillar_features'] (pillar特征 [n,64])
2023-05-12 opencood/models/point_pillar.py
pointpillar模型的整体架构,添加了论文中对于模型组成的解释。
2023-05-10 opencood/tools/train.py
模型的训练流程
所有的模型都使用这一训练流程
2023-05-09 opencood/hypes_yaml/point_pillar_early_fusion.yaml
配置文件:包含数据集,前处理,数据增强,融合方法,模型,后处理,损失函数,优化器,学习率调度器。
其他的配置文件和此结构相同