Skip to content

GitLab

Unverified Commit a991105c authored by Shaoshuai Shi's avatar Shaoshuai Shi Committed by GitHub
Browse files

Release the codes of PV-RCNN++, update OpenPCDet to v0.5.2 

Release the codes of PV-RCNN++, update OpenPCDet to v0.5.2
parents 1483517a b6fbf07f
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 1317 additions and 1 deletion
pcdet/ops/pointnet2/pointnet2_stack/src/vector_pool.cpp 0 → 100644
View file @ a991105c
/*
Vector-pool aggregation based local feature aggregation for point cloud.
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection
https://arxiv.org/abs/2102.00463
Written by Shaoshuai Shi
All Rights Reserved 2020.
*/
#include <torch/serialize/tensor.h>
#include <vector>
#include <THC/THC.h>
#include <cuda.h>
#include <cuda_runtime_api.h>
#include "vector_pool_gpu.h"
extern THCState *state;
#define CHECK_CUDA(x) do { \
if (!x.type().is_cuda()) { \
fprintf(stderr, "%s must be CUDA tensor at %s:%d\n", #x, __FILE__, __LINE__); \
exit(-1); \
} \
} while (0)
#define CHECK_CONTIGUOUS(x) do { \
if (!x.is_contiguous()) { \
fprintf(stderr, "%s must be contiguous tensor at %s:%d\n", #x, __FILE__, __LINE__); \
exit(-1); \
} \
} while (0)
#define CHECK_INPUT(x) CHECK_CUDA(x);CHECK_CONTIGUOUS(x)
int query_stacked_local_neighbor_idxs_wrapper_stack(at::Tensor support_xyz_tensor, at::Tensor xyz_batch_cnt_tensor,
at::Tensor new_xyz_tensor, at::Tensor new_xyz_batch_cnt_tensor,
at::Tensor stack_neighbor_idxs_tensor, at::Tensor start_len_tensor, at::Tensor cumsum_tensor,
int avg_length_of_neighbor_idxs, float max_neighbour_distance, int nsample, int neighbor_type){
// support_xyz: (N1 + N2 ..., 3) xyz coordinates of the features
// xyz_batch_cnt: (batch_size), [N1, N2, ...]
// new_xyz: (M1 + M2 ..., 3) centers of the ball query
// new_xyz_grid_centers: (M1 + M2 ..., num_total_grids, 3) grids centers of each grid
// new_xyz_batch_cnt: (batch_size), [M1, M2, ...]
// new_xyz_grid_idxs: (M1 + M2 ..., num_total_grids, 3) three-nn
// new_xyz_grid_dist2: (M1 + M2 ..., num_total_grids, 3) square of dist of three-nn
// num_grid_x, num_grid_y, num_grid_z: number of grids in each local area centered at new_xyz
// nsample: find all (-1), find limited number(>0)
// neighbor_type: 1: ball, others: cube
CHECK_INPUT(support_xyz_tensor);
CHECK_INPUT(xyz_batch_cnt_tensor);
CHECK_INPUT(new_xyz_tensor);
CHECK_INPUT(new_xyz_batch_cnt_tensor);
CHECK_INPUT(stack_neighbor_idxs_tensor);
CHECK_INPUT(start_len_tensor);
CHECK_INPUT(cumsum_tensor);
const float *support_xyz = support_xyz_tensor.data<float>();
const int *xyz_batch_cnt = xyz_batch_cnt_tensor.data<int>();
const float *new_xyz = new_xyz_tensor.data<float>();
const int *new_xyz_batch_cnt = new_xyz_batch_cnt_tensor.data<int>();
int *stack_neighbor_idxs = stack_neighbor_idxs_tensor.data<int>();
int *start_len = start_len_tensor.data<int>();
int *cumsum = cumsum_tensor.data<int>();
int batch_size = xyz_batch_cnt_tensor.size(0);
int M = new_xyz_tensor.size(0);
query_stacked_local_neighbor_idxs_kernel_launcher_stack(
support_xyz, xyz_batch_cnt, new_xyz, new_xyz_batch_cnt,
stack_neighbor_idxs, start_len, cumsum, avg_length_of_neighbor_idxs,
max_neighbour_distance, batch_size, M, nsample, neighbor_type
);
return 0;
}
int query_three_nn_by_stacked_local_idxs_wrapper_stack(at::Tensor support_xyz_tensor,
at::Tensor new_xyz_tensor, at::Tensor new_xyz_grid_centers_tensor,
at::Tensor new_xyz_grid_idxs_tensor, at::Tensor new_xyz_grid_dist2_tensor,
at::Tensor stack_neighbor_idxs_tensor, at::Tensor start_len_tensor,
int M, int num_total_grids){
// support_xyz: (N1 + N2 ..., 3) xyz coordinates of the features
// new_xyz: (M1 + M2 ..., 3) centers of the ball query
// new_xyz_grid_centers: (M1 + M2 ..., num_total_grids, 3) grids centers of each grid
// new_xyz_grid_idxs: (M1 + M2 ..., num_total_grids, 3) three-nn
// new_xyz_grid_dist2: (M1 + M2 ..., num_total_grids, 3) square of dist of three-nn
// stack_neighbor_idxs: (max_length_of_neighbor_idxs)
// start_len: (M1 + M2, 2) [start_offset, neighbor_length]
CHECK_INPUT(support_xyz_tensor);
CHECK_INPUT(new_xyz_tensor);
CHECK_INPUT(new_xyz_grid_centers_tensor);
CHECK_INPUT(new_xyz_grid_idxs_tensor);
CHECK_INPUT(new_xyz_grid_dist2_tensor);
CHECK_INPUT(stack_neighbor_idxs_tensor);
CHECK_INPUT(start_len_tensor);
const float *support_xyz = support_xyz_tensor.data<float>();
const float *new_xyz = new_xyz_tensor.data<float>();
const float *new_xyz_grid_centers = new_xyz_grid_centers_tensor.data<float>();
int *new_xyz_grid_idxs = new_xyz_grid_idxs_tensor.data<int>();
float *new_xyz_grid_dist2 = new_xyz_grid_dist2_tensor.data<float>();
int *stack_neighbor_idxs = stack_neighbor_idxs_tensor.data<int>();
int *start_len = start_len_tensor.data<int>();
query_three_nn_by_stacked_local_idxs_kernel_launcher_stack(
support_xyz, new_xyz, new_xyz_grid_centers,
new_xyz_grid_idxs, new_xyz_grid_dist2, stack_neighbor_idxs, start_len,
M, num_total_grids
);
return 0;
}
int vector_pool_wrapper_stack(at::Tensor support_xyz_tensor, at::Tensor xyz_batch_cnt_tensor,
at::Tensor support_features_tensor, at::Tensor new_xyz_tensor, at::Tensor new_xyz_batch_cnt_tensor,
at::Tensor new_features_tensor, at::Tensor new_local_xyz_tensor,
at::Tensor point_cnt_of_grid_tensor, at::Tensor grouped_idxs_tensor,
int num_grid_x, int num_grid_y, int num_grid_z, float max_neighbour_distance, int use_xyz,
int num_max_sum_points, int nsample, int neighbor_type, int pooling_type){
// support_xyz_tensor: (N1 + N2 ..., 3) xyz coordinates of the features
// support_features_tensor: (N1 + N2 ..., C)
// xyz_batch_cnt: (batch_size), [N1, N2, ...]
// new_xyz_tensor: (M1 + M2 ..., 3) centers of new positions
// new_features_tensor: (M1 + M2 ..., C)
// new_xyz_batch_cnt: (batch_size), [M1, M2, ...]
// point_cnt_of_grid: (M1 + M2 ..., num_total_grids)
// grouped_idxs_tensor: (num_max_sum_points, 3)
// num_grid_x, num_grid_y, num_grid_z: number of grids in each local area centered at new_xyz
// use_xyz: whether to calculate new_local_xyz
// neighbor_type: 1: ball, others: cube
// pooling_type: 0: avg_pool, 1: random choice
CHECK_INPUT(support_xyz_tensor);
CHECK_INPUT(support_features_tensor);
CHECK_INPUT(xyz_batch_cnt_tensor);
CHECK_INPUT(new_xyz_tensor);
CHECK_INPUT(new_xyz_batch_cnt_tensor);
CHECK_INPUT(new_features_tensor);
CHECK_INPUT(new_local_xyz_tensor);
CHECK_INPUT(point_cnt_of_grid_tensor);
CHECK_INPUT(grouped_idxs_tensor);
const float *support_xyz = support_xyz_tensor.data<float>();
const float *support_features = support_features_tensor.data<float>();
const int *xyz_batch_cnt = xyz_batch_cnt_tensor.data<int>();
const float *new_xyz = new_xyz_tensor.data<float>();
const int *new_xyz_batch_cnt = new_xyz_batch_cnt_tensor.data<int>();
float *new_features = new_features_tensor.data<float>();
float *new_local_xyz = new_local_xyz_tensor.data<float>();
int *point_cnt_of_grid = point_cnt_of_grid_tensor.data<int>();
int *grouped_idxs = grouped_idxs_tensor.data<int>();
int N = support_xyz_tensor.size(0);
int batch_size = xyz_batch_cnt_tensor.size(0);
int M = new_xyz_tensor.size(0);
int num_c_out = new_features_tensor.size(1);
int num_c_in = support_features_tensor.size(1);
int num_total_grids = point_cnt_of_grid_tensor.size(1);
int cum_sum = vector_pool_kernel_launcher_stack(
support_xyz, support_features, xyz_batch_cnt,
new_xyz, new_features, new_local_xyz, new_xyz_batch_cnt,
point_cnt_of_grid, grouped_idxs,
num_grid_x, num_grid_y, num_grid_z, max_neighbour_distance,
batch_size, N, M, num_c_in, num_c_out, num_total_grids, use_xyz, num_max_sum_points, nsample, neighbor_type, pooling_type
);
return cum_sum;
}
int vector_pool_grad_wrapper_stack(at::Tensor grad_new_features_tensor,
at::Tensor point_cnt_of_grid_tensor, at::Tensor grouped_idxs_tensor,
at::Tensor grad_support_features_tensor) {
// grad_new_features_tensor: (M1 + M2 ..., C_out)
// point_cnt_of_grid_tensor: (M1 + M2 ..., num_total_grids)
// grouped_idxs_tensor: (num_max_sum_points, 3) [idx of support_xyz, idx of new_xyz, idx of grid_idx in new_xyz]
// grad_support_features_tensor: (N1 + N2 ..., C_in)
CHECK_INPUT(grad_new_features_tensor);
CHECK_INPUT(point_cnt_of_grid_tensor);
CHECK_INPUT(grouped_idxs_tensor);
CHECK_INPUT(grad_support_features_tensor);
int M = grad_new_features_tensor.size(0);
int num_c_out = grad_new_features_tensor.size(1);
int N = grad_support_features_tensor.size(0);
int num_c_in = grad_support_features_tensor.size(1);
int num_total_grids = point_cnt_of_grid_tensor.size(1);
int num_max_sum_points = grouped_idxs_tensor.size(0);
const float *grad_new_features = grad_new_features_tensor.data<float>();
const int *point_cnt_of_grid = point_cnt_of_grid_tensor.data<int>();
const int *grouped_idxs = grouped_idxs_tensor.data<int>();
float *grad_support_features = grad_support_features_tensor.data<float>();
vector_pool_grad_kernel_launcher_stack(
grad_new_features, point_cnt_of_grid, grouped_idxs, grad_support_features,
N, M, num_c_out, num_c_in, num_total_grids, num_max_sum_points
);
return 1;
}
pcdet/ops/pointnet2/pointnet2_stack/src/vector_pool_gpu.cu 0 → 100644
View file @ a991105c
This diff is collapsed.
pcdet/ops/pointnet2/pointnet2_stack/src/vector_pool_gpu.h 0 → 100644
View file @ a991105c
/*
Vector-pool aggregation based local feature aggregation for point cloud.
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection
https://arxiv.org/abs/2102.00463
Written by Shaoshuai Shi
All Rights Reserved 2020.
*/
#ifndef _STACK_VECTOR_POOL_GPU_H
#define _STACK_VECTOR_POOL_GPU_H
#include <torch/serialize/tensor.h>
#include <vector>
#include <cuda.h>
#include <cuda_runtime_api.h>
int query_stacked_local_neighbor_idxs_kernel_launcher_stack(
const float *support_xyz, const int *xyz_batch_cnt, const float *new_xyz, const int *new_xyz_batch_cnt,
int *stack_neighbor_idxs, int *start_len, int *cumsum, int avg_length_of_neighbor_idxs,
float max_neighbour_distance, int batch_size, int M, int nsample, int neighbor_type);
int query_stacked_local_neighbor_idxs_wrapper_stack(at::Tensor support_xyz_tensor, at::Tensor xyz_batch_cnt_tensor,
at::Tensor new_xyz_tensor, at::Tensor new_xyz_batch_cnt_tensor,
at::Tensor stack_neighbor_idxs_tensor, at::Tensor start_len_tensor, at::Tensor cumsum_tensor,
int avg_length_of_neighbor_idxs, float max_neighbour_distance, int nsample, int neighbor_type);
int query_three_nn_by_stacked_local_idxs_kernel_launcher_stack(
const float *support_xyz, const float *new_xyz, const float *new_xyz_grid_centers,
int *new_xyz_grid_idxs, float *new_xyz_grid_dist2,
const int *stack_neighbor_idxs, const int *start_len,
int M, int num_total_grids);
int query_three_nn_by_stacked_local_idxs_wrapper_stack(at::Tensor support_xyz_tensor,
at::Tensor new_xyz_tensor, at::Tensor new_xyz_grid_centers_tensor,
at::Tensor new_xyz_grid_idxs_tensor, at::Tensor new_xyz_grid_dist2_tensor,
at::Tensor stack_neighbor_idxs_tensor, at::Tensor start_len_tensor,
int M, int num_total_grids);
int vector_pool_wrapper_stack(at::Tensor support_xyz_tensor, at::Tensor xyz_batch_cnt_tensor,
at::Tensor support_features_tensor, at::Tensor new_xyz_tensor, at::Tensor new_xyz_batch_cnt_tensor,
at::Tensor new_features_tensor, at::Tensor new_local_xyz,
at::Tensor point_cnt_of_grid_tensor, at::Tensor grouped_idxs_tensor,
int num_grid_x, int num_grid_y, int num_grid_z, float max_neighbour_distance, int use_xyz,
int num_max_sum_points, int nsample, int neighbor_type, int pooling_type);
int vector_pool_kernel_launcher_stack(
const float *support_xyz, const float *support_features, const int *xyz_batch_cnt,
const float *new_xyz, float *new_features, float * new_local_xyz, const int *new_xyz_batch_cnt,
int *point_cnt_of_grid, int *grouped_idxs,
int num_grid_x, int num_grid_y, int num_grid_z, float max_neighbour_distance,
int batch_size, int N, int M, int num_c_in, int num_c_out, int num_total_grids, int use_xyz,
int num_max_sum_points, int nsample, int neighbor_type, int pooling_type);
int vector_pool_grad_wrapper_stack(at::Tensor grad_new_features_tensor,
at::Tensor point_cnt_of_grid_tensor, at::Tensor grouped_idxs_tensor,
at::Tensor grad_support_features_tensor);
void vector_pool_grad_kernel_launcher_stack(
const float *grad_new_features, const int *point_cnt_of_grid, const int *grouped_idxs,
float *grad_support_features, int N, int M, int num_c_out, int num_c_in, int num_total_grids,
int num_max_sum_points);
#endif
setup.py
View file @ a991105c
......@@ -28,7 +28,7 @@ def write_version_to_file(version, target_file):
if __name__ == '__main__':
version = '0.5.1+%s' % get_git_commit_number()
version = '0.5.2+%s' % get_git_commit_number()
write_version_to_file(version, 'pcdet/version.py')
setup(
......@@ -97,6 +97,8 @@ if __name__ == '__main__':
'src/interpolate_gpu.cu',
'src/voxel_query.cpp',
'src/voxel_query_gpu.cu',
'src/vector_pool.cpp',
'src/vector_pool_gpu.cu'
],
),
make_cuda_ext(
......
tools/cfgs/waymo_models/pv_rcnn_plusplus.yaml 0 → 100644
View file @ a991105c
CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']
DATA_CONFIG:
_BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml
MODEL:
NAME: PVRCNNPlusPlus
VFE:
NAME: MeanVFE
BACKBONE_3D:
NAME: VoxelBackBone8x
MAP_TO_BEV:
NAME: HeightCompression
NUM_BEV_FEATURES: 256
BACKBONE_2D:
NAME: BaseBEVBackbone
LAYER_NUMS: [5, 5]
LAYER_STRIDES: [1, 2]
NUM_FILTERS: [128, 256]
UPSAMPLE_STRIDES: [1, 2]
NUM_UPSAMPLE_FILTERS: [256, 256]
DENSE_HEAD:
NAME: CenterHead
CLASS_AGNOSTIC: False
CLASS_NAMES_EACH_HEAD: [
[ 'Vehicle', 'Pedestrian', 'Cyclist' ]
]
SHARED_CONV_CHANNEL: 64
USE_BIAS_BEFORE_NORM: True
NUM_HM_CONV: 2
SEPARATE_HEAD_CFG:
HEAD_ORDER: [ 'center', 'center_z', 'dim', 'rot' ]
HEAD_DICT: {
'center': { 'out_channels': 2, 'num_conv': 2 },
'center_z': { 'out_channels': 1, 'num_conv': 2 },
'dim': { 'out_channels': 3, 'num_conv': 2 },
'rot': { 'out_channels': 2, 'num_conv': 2 },
}
TARGET_ASSIGNER_CONFIG:
FEATURE_MAP_STRIDE: 8
NUM_MAX_OBJS: 500
GAUSSIAN_OVERLAP: 0.1
MIN_RADIUS: 2
LOSS_CONFIG:
LOSS_WEIGHTS: {
'cls_weight': 1.0,
'loc_weight': 2.0,
'code_weights': [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
}
POST_PROCESSING:
SCORE_THRESH: 0.1
POST_CENTER_LIMIT_RANGE: [ -75.2, -75.2, -2, 75.2, 75.2, 4 ]
MAX_OBJ_PER_SAMPLE: 500
NMS_CONFIG:
NMS_TYPE: nms_gpu
NMS_THRESH: 0.7
NMS_PRE_MAXSIZE: 4096
NMS_POST_MAXSIZE: 500
PFE:
NAME: VoxelSetAbstraction
POINT_SOURCE: raw_points
NUM_KEYPOINTS: 4096
NUM_OUTPUT_FEATURES: 90
SAMPLE_METHOD: SPC
SPC_SAMPLING:
NUM_SECTORS: 6
SAMPLE_RADIUS_WITH_ROI: 1.6
FEATURES_SOURCE: ['bev', 'x_conv3', 'x_conv4', 'raw_points']
SA_LAYER:
raw_points:
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 2
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 32 ]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 2.4
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 2, 2, 2 ]
MAX_NEIGHBOR_DISTANCE: 0.2
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 32, 32 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 0.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 32, 32 ]
x_conv3:
DOWNSAMPLE_FACTOR: 4
INPUT_CHANNELS: 64
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 32
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [128]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 4.0
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [3, 3, 3]
MAX_NEIGHBOR_DISTANCE: 1.2
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [64, 64]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 2.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
x_conv4:
DOWNSAMPLE_FACTOR: 8
INPUT_CHANNELS: 64
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 32
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 128 ]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 6.4
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 2.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 4.8
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
POINT_HEAD:
NAME: PointHeadSimple
CLS_FC: [256, 256]
CLASS_AGNOSTIC: True
USE_POINT_FEATURES_BEFORE_FUSION: True
TARGET_CONFIG:
GT_EXTRA_WIDTH: [0.2, 0.2, 0.2]
LOSS_CONFIG:
LOSS_REG: smooth-l1
LOSS_WEIGHTS: {
'point_cls_weight': 1.0,
}
ROI_HEAD:
NAME: PVRCNNHead
CLASS_AGNOSTIC: True
SHARED_FC: [256, 256]
CLS_FC: [256, 256]
REG_FC: [256, 256]
DP_RATIO: 0.3
NMS_CONFIG:
TRAIN:
NMS_TYPE: nms_gpu
MULTI_CLASSES_NMS: False
NMS_PRE_MAXSIZE: 9000
NMS_POST_MAXSIZE: 512
NMS_THRESH: 0.8
TEST:
NMS_TYPE: nms_gpu
MULTI_CLASSES_NMS: False
NMS_PRE_MAXSIZE: 1024
NMS_POST_MAXSIZE: 100
NMS_THRESH: 0.7
SCORE_THRESH: 0.1
# NMS_PRE_MAXSIZE: 4096
# NMS_POST_MAXSIZE: 500
# NMS_THRESH: 0.85
ROI_GRID_POOL:
GRID_SIZE: 6
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: voxel_random_choice
NUM_REDUCED_CHANNELS: 30
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 128 ]
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 0.8
NEIGHBOR_NSAMPLE: 32
POST_MLPS: [ 64, 64 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 1.6
NEIGHBOR_NSAMPLE: 32
POST_MLPS: [ 64, 64 ]
TARGET_CONFIG:
BOX_CODER: ResidualCoder
ROI_PER_IMAGE: 128
FG_RATIO: 0.5
SAMPLE_ROI_BY_EACH_CLASS: True
CLS_SCORE_TYPE: roi_iou
CLS_FG_THRESH: 0.75
CLS_BG_THRESH: 0.25
CLS_BG_THRESH_LO: 0.1
HARD_BG_RATIO: 0.8
REG_FG_THRESH: 0.55
LOSS_CONFIG:
CLS_LOSS: BinaryCrossEntropy
REG_LOSS: smooth-l1
CORNER_LOSS_REGULARIZATION: True
LOSS_WEIGHTS: {
'rcnn_cls_weight': 1.0,
'rcnn_reg_weight': 1.0,
'rcnn_corner_weight': 1.0,
'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
}
POST_PROCESSING:
RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
SCORE_THRESH: 0.1
OUTPUT_RAW_SCORE: False
EVAL_METRIC: waymo
NMS_CONFIG:
MULTI_CLASSES_NMS: False
NMS_TYPE: nms_gpu
NMS_THRESH: 0.7
NMS_PRE_MAXSIZE: 4096
NMS_POST_MAXSIZE: 500
OPTIMIZATION:
BATCH_SIZE_PER_GPU: 2
NUM_EPOCHS: 30
OPTIMIZER: adam_onecycle
LR: 0.01
WEIGHT_DECAY: 0.001
MOMENTUM: 0.9
MOMS: [0.95, 0.85]
PCT_START: 0.4
DIV_FACTOR: 10
DECAY_STEP_LIST: [35, 45]
LR_DECAY: 0.1
LR_CLIP: 0.0000001
LR_WARMUP: False
WARMUP_EPOCH: 1
GRAD_NORM_CLIP: 10
\ No newline at end of file
tools/cfgs/waymo_models/pv_rcnn_plusplus_resnet.yaml 0 → 100644
View file @ a991105c
CLASS_NAMES: ['Vehicle', 'Pedestrian', 'Cyclist']
DATA_CONFIG:
_BASE_CONFIG_: cfgs/dataset_configs/waymo_dataset.yaml
MODEL:
NAME: PVRCNNPlusPlus
VFE:
NAME: MeanVFE
BACKBONE_3D:
NAME: VoxelResBackBone8x
MAP_TO_BEV:
NAME: HeightCompression
NUM_BEV_FEATURES: 256
BACKBONE_2D:
NAME: BaseBEVBackbone
LAYER_NUMS: [5, 5]
LAYER_STRIDES: [1, 2]
NUM_FILTERS: [128, 256]
UPSAMPLE_STRIDES: [1, 2]
NUM_UPSAMPLE_FILTERS: [256, 256]
DENSE_HEAD:
NAME: CenterHead
CLASS_AGNOSTIC: False
CLASS_NAMES_EACH_HEAD: [
[ 'Vehicle', 'Pedestrian', 'Cyclist' ]
]
SHARED_CONV_CHANNEL: 64
USE_BIAS_BEFORE_NORM: True
NUM_HM_CONV: 2
SEPARATE_HEAD_CFG:
HEAD_ORDER: [ 'center', 'center_z', 'dim', 'rot' ]
HEAD_DICT: {
'center': { 'out_channels': 2, 'num_conv': 2 },
'center_z': { 'out_channels': 1, 'num_conv': 2 },
'dim': { 'out_channels': 3, 'num_conv': 2 },
'rot': { 'out_channels': 2, 'num_conv': 2 },
}
TARGET_ASSIGNER_CONFIG:
FEATURE_MAP_STRIDE: 8
NUM_MAX_OBJS: 500
GAUSSIAN_OVERLAP: 0.1
MIN_RADIUS: 2
LOSS_CONFIG:
LOSS_WEIGHTS: {
'cls_weight': 1.0,
'loc_weight': 2.0,
'code_weights': [ 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 ]
}
POST_PROCESSING:
SCORE_THRESH: 0.1
POST_CENTER_LIMIT_RANGE: [ -75.2, -75.2, -2, 75.2, 75.2, 4 ]
MAX_OBJ_PER_SAMPLE: 500
NMS_CONFIG:
NMS_TYPE: nms_gpu
NMS_THRESH: 0.7
NMS_PRE_MAXSIZE: 4096
NMS_POST_MAXSIZE: 500
PFE:
NAME: VoxelSetAbstraction
POINT_SOURCE: raw_points
NUM_KEYPOINTS: 4096
NUM_OUTPUT_FEATURES: 90
SAMPLE_METHOD: SPC
SPC_SAMPLING:
NUM_SECTORS: 6
SAMPLE_RADIUS_WITH_ROI: 1.6
FEATURES_SOURCE: ['bev', 'x_conv3', 'x_conv4', 'raw_points']
SA_LAYER:
raw_points:
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 2
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 32 ]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 2.4
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 2, 2, 2 ]
MAX_NEIGHBOR_DISTANCE: 0.2
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 32, 32 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 0.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 32, 32 ]
x_conv3:
DOWNSAMPLE_FACTOR: 4
INPUT_CHANNELS: 64
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 32
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [128]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 4.0
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [3, 3, 3]
MAX_NEIGHBOR_DISTANCE: 1.2
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [64, 64]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 2.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
x_conv4:
DOWNSAMPLE_FACTOR: 8
INPUT_CHANNELS: 64
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: local_interpolation
NUM_REDUCED_CHANNELS: 32
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 128 ]
FILTER_NEIGHBOR_WITH_ROI: True
RADIUS_OF_NEIGHBOR_WITH_ROI: 6.4
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 2.4
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 4.8
NEIGHBOR_NSAMPLE: -1
POST_MLPS: [ 64, 64 ]
POINT_HEAD:
NAME: PointHeadSimple
CLS_FC: [256, 256]
CLASS_AGNOSTIC: True
USE_POINT_FEATURES_BEFORE_FUSION: True
TARGET_CONFIG:
GT_EXTRA_WIDTH: [0.2, 0.2, 0.2]
LOSS_CONFIG:
LOSS_REG: smooth-l1
LOSS_WEIGHTS: {
'point_cls_weight': 1.0,
}
ROI_HEAD:
NAME: PVRCNNHead
CLASS_AGNOSTIC: True
SHARED_FC: [256, 256]
CLS_FC: [256, 256]
REG_FC: [256, 256]
DP_RATIO: 0.3
NMS_CONFIG:
TRAIN:
NMS_TYPE: nms_gpu
MULTI_CLASSES_NMS: False
NMS_PRE_MAXSIZE: 9000
NMS_POST_MAXSIZE: 512
NMS_THRESH: 0.8
TEST:
NMS_TYPE: nms_gpu
MULTI_CLASSES_NMS: False
NMS_PRE_MAXSIZE: 1024
NMS_POST_MAXSIZE: 100
NMS_THRESH: 0.7
SCORE_THRESH: 0.1
# NMS_PRE_MAXSIZE: 4096
# NMS_POST_MAXSIZE: 500
# NMS_THRESH: 0.85
ROI_GRID_POOL:
GRID_SIZE: 6
NAME: VectorPoolAggregationModuleMSG
NUM_GROUPS: 2
LOCAL_AGGREGATION_TYPE: voxel_random_choice
NUM_REDUCED_CHANNELS: 30
NUM_CHANNELS_OF_LOCAL_AGGREGATION: 32
MSG_POST_MLPS: [ 128 ]
GROUP_CFG_0:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 0.8
NEIGHBOR_NSAMPLE: 32
POST_MLPS: [ 64, 64 ]
GROUP_CFG_1:
NUM_LOCAL_VOXEL: [ 3, 3, 3 ]
MAX_NEIGHBOR_DISTANCE: 1.6
NEIGHBOR_NSAMPLE: 32
POST_MLPS: [ 64, 64 ]
TARGET_CONFIG:
BOX_CODER: ResidualCoder
ROI_PER_IMAGE: 128
FG_RATIO: 0.5
SAMPLE_ROI_BY_EACH_CLASS: True
CLS_SCORE_TYPE: roi_iou
CLS_FG_THRESH: 0.75
CLS_BG_THRESH: 0.25
CLS_BG_THRESH_LO: 0.1
HARD_BG_RATIO: 0.8
REG_FG_THRESH: 0.55
LOSS_CONFIG:
CLS_LOSS: BinaryCrossEntropy
REG_LOSS: smooth-l1
CORNER_LOSS_REGULARIZATION: True
LOSS_WEIGHTS: {
'rcnn_cls_weight': 1.0,
'rcnn_reg_weight': 1.0,
'rcnn_corner_weight': 1.0,
'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
}
POST_PROCESSING:
RECALL_THRESH_LIST: [0.3, 0.5, 0.7]
SCORE_THRESH: 0.1
OUTPUT_RAW_SCORE: False
EVAL_METRIC: waymo
NMS_CONFIG:
MULTI_CLASSES_NMS: False
NMS_TYPE: nms_gpu
NMS_THRESH: 0.7
NMS_PRE_MAXSIZE: 4096
NMS_POST_MAXSIZE: 500
OPTIMIZATION:
BATCH_SIZE_PER_GPU: 2
NUM_EPOCHS: 30
OPTIMIZER: adam_onecycle
LR: 0.01
WEIGHT_DECAY: 0.001
MOMENTUM: 0.9
MOMS: [0.95, 0.85]
PCT_START: 0.4
DIV_FACTOR: 10
DECAY_STEP_LIST: [35, 45]
LR_DECAY: 0.1
LR_CLIP: 0.0000001
LR_WARMUP: False
WARMUP_EPOCH: 1
GRAD_NORM_CLIP: 10
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment