SA-SSD icon indicating copy to clipboard operation
SA-SSD copied to clipboard

SA-SSD: Structure Aware Single-stage 3D Object Detection from Point Cloud (CVPR 2020)

Results 57 SA-SSD issues
Sort by recently updated
recently updated
newest added

The **model configs** are mentioned as follow in file config/multi_cfg.py : model = dict( type='SingleStageDetector', backbone=dict( type='SimpleVoxel', num_input_features=4, use_norm=True, num_filters=[32, 64], with_distance=False), neck=dict( type='SpMiddleFHD', output_shape=[150,440,440], num_input_features=4, num_hidden_features=64 * 19, ),...

Are there any advices on how to train and inference on a customize dataset with lidar data only?

@skyhehe123 I am trying to visualize the results after performing training. However I do not see any code-instructions to do so. Can you please point me towards how you visualized...

FileNotFoundError: file "/home/xys/SA-SSD-master/tools/config" does not exist

I download the pretrained model, and evaluate the model: cd tools python3 test.py ../configs/car_cfg.py ../saved_model_vehicle/epoch_50.pth Then the error occur: ==> Loading parameters from checkpoint ../saved_model_vehicle/epoch_50.pth to GPU Traceback (most recent...

When I run the create_data.py, the error always occurs: ImportError: /home/weijk/ws/SA-SSD-master/mmdet/ops/points_op/points_op_cpu.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZN6caffe28TypeMeta21_typeMetaDataInstanceIN3c108BFloat16EEEPKNS_6detail12TypeMetaDataEv with cudatoolkit-9.0, pytorch-1.1.0, torchvision-0.3.0, torch-1.1.0

/media/adas/File/wdx/dianyun/SA-SSD/mmdet/core/bbox3d/geometry.py:189: NumbaWarning: Compilation is falling back to object mode WITH looplifting enabled because Function "points_in_convex_polygon_3d_jit" failed type inference due to: Invalid use of type(CPUDispatcher()) with parameters (array(float32, 4d, A)) *...

I am not able to train the model, getting error related to CUDA. ``` python train.py configs/car_cfg.py ``` ``` 2020-09-29 16:35:43,686 - INFO - Distributed training: False 2020-09-29 16:35:43,686 -...

Hello, skyhehe123! Thanks for you excellent work. In your paper, I found the pre-processing method of point cloud is different from voxelization in SECOND method and is more efficient and...