thebears/deepstream_yolo
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add PP-YOLOE support

Browse Source
This commit is contained in:
Marcos Luciano
2022-07-24 18:00:47 -03:00
parent d09879d557
commit a3782ed65e
51 changed files with 1812 additions and 600 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
nvdsinfer_custom_impl_Yolo/sortDetections.cu
View File

@@ -37,13 +37,11 @@ cudaError_t sortDetections(
float* _d_scores = reinterpret_cast<float*>(d_scores) + (batch * outputSize);
int* _countData = reinterpret_cast<int*>(countData) + (batch);
int* _count = (int*)malloc(sizeof(int));
cudaMemcpy(_count, (int*)&_countData[0], sizeof(int), cudaMemcpyDeviceToHost);
int count = _count[0];
int count;
cudaMemcpy(&count, _countData, sizeof(int), cudaMemcpyDeviceToHost);
if (count == 0)
{
free(_count);
return cudaGetLastError();
}
@@ -72,13 +70,13 @@ cudaError_t sortDetections(
int _topK = count < topK ? count : topK;
int threads_per_block = 0;
int number_of_blocks = 4;
int threads_per_block = 16;
int number_of_blocks = 0;
if (_topK % 2 == 0 && _topK >= number_of_blocks)
threads_per_block = _topK / number_of_blocks;
if (_topK % 2 == 0 && _topK >= threads_per_block)
number_of_blocks = _topK / threads_per_block;
else
threads_per_block = (_topK / number_of_blocks) + 1;
number_of_blocks = (_topK / threads_per_block) + 1;
sortOutput<<<number_of_blocks, threads_per_block, 0, stream>>>(
_d_indexes, _d_scores, reinterpret_cast<float*>(d_boxes) + (batch * 4 * outputSize),
@@ -89,8 +87,6 @@ cudaError_t sortDetections(
cudaFree(d_keys_out);
cudaFree(d_values_out);
cudaFree(d_temp_storage);
free(_count);
}
return cudaGetLastError();
}