haotian/duan8
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix multi-batch problem in int8 quantization (#1604)

Browse Source 
* Use the configuration of FP16 by default.

* Fix multi-batch problem in int8 quantization
This commit is contained in:
mpj1234 2024-12-05 14:45:24 +08:00 committed by GitHub
parent b34d799da1
commit 62c16801b4
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
yolo11/src/model.cpp
View File

@ -116,7 +116,7 @@ nvinfer1::IHostMemory* buildEngineYolo11Cls(nvinfer1::IBuilder* builder, nvinfer
std::cout << "Your platform supports int8: " << (builder->platformHasFastInt8() ? "true" : "false") << std::endl;
assert(builder->platformHasFastInt8());
config->setFlag(nvinfer1::BuilderFlag::kINT8);
auto* calibrator = new Int8EntropyCalibrator2(1, kClsInputW, kClsInputH, kInputQuantizationFolder,
auto* calibrator = new Int8EntropyCalibrator2(kBatchSize, kClsInputW, kClsInputH, kInputQuantizationFolder,
"int8calib.table", kInputTensorName);
config->setInt8Calibrator(calibrator);
#endif
@ -392,8 +392,8 @@ nvinfer1::IHostMemory* buildEngineYolo11Det(nvinfer1::IBuilder* builder, nvinfer
std::cout << "Your platform support int8: " << (builder->platformHasFastInt8() ? "true" : "false") << std::endl;
assert(builder->platformHasFastInt8());
config->setFlag(nvinfer1::BuilderFlag::kINT8);
auto* calibrator = new Int8EntropyCalibrator2(1, kInputW, kInputH, kInputQuantizationFolder, "int8calib.table",
kInputTensorName);
auto* calibrator = new Int8EntropyCalibrator2(kBatchSize, kInputW, kInputH, kInputQuantizationFolder,
"int8calib.table", kInputTensorName);
config->setInt8Calibrator(calibrator);
#endif
@ -775,8 +775,8 @@ nvinfer1::IHostMemory* buildEngineYolo11Seg(nvinfer1::IBuilder* builder, nvinfer
std::cout << "Your platform support int8: " << (builder->platformHasFastInt8() ? "true" : "false") << std::endl;
assert(builder->platformHasFastInt8());
config->setFlag(nvinfer1::BuilderFlag::kINT8);
auto* calibrator = new Int8EntropyCalibrator2(1, kInputW, kInputH, kInputQuantizationFolder, "int8calib.table",
kInputTensorName);
auto* calibrator = new Int8EntropyCalibrator2(kBatchSize, kInputW, kInputH, kInputQuantizationFolder,
"int8calib.table", kInputTensorName);
config->setInt8Calibrator(calibrator);
#endif
@ -1066,8 +1066,8 @@ nvinfer1::IHostMemory* buildEngineYolo11Pose(nvinfer1::IBuilder* builder, nvinfe
std::cout << "Your platform support int8: " << (builder->platformHasFastInt8() ? "true" : "false") << std::endl;
assert(builder->platformHasFastInt8());
config->setFlag(nvinfer1::BuilderFlag::kINT8);
auto* calibrator = new Int8EntropyCalibrator2(1, kInputW, kInputH, kInputQuantizationFolder, "int8calib.table",
kInputTensorName);
auto* calibrator = new Int8EntropyCalibrator2(kBatchSize, kInputW, kInputH, kInputQuantizationFolder,
"int8calib.table", kInputTensorName);
config->setInt8Calibrator(calibrator);
#endif