Hello,
04_video_dec_trt example
How do I use it with yolov3 (convert to caffemodel after learning using darknet), not resnet?
Thank you.
Hello,
04_video_dec_trt example
How do I use it with yolov3 (convert to caffemodel after learning using darknet), not resnet?
Thank you.
Hi,
1. Please modify the model name first.
const char *GOOGLE_NET_DEPLOY_NAME =
"../../data/Model/GoogleNet_one_class/GoogleNet_modified_oneClass_halfHD.prototxt";
const char *GOOGLE_NET_MODEL_NAME =
"../../data/Model/GoogleNet_one_class/GoogleNet_modified_oneClass_halfHD.caffemodel";
2. Add the model definition with g_pModelNetAttr in common/algorithm/trt/trt_inference.h:
struct {
const int classCnt;
float THRESHOLD[3];
const char *INPUT_BLOB_NAME;
const char *OUTPUT_BLOB_NAME;
const char *OUTPUT_BBOX_NAME;
const int STRIDE;
const int WORKSPACE_SIZE;
int offsets[3];
float input_scale[3];
float bbox_output_scales[4];
const int ParseFunc_ID;
} *g_pModelNetAttr, gModelNetAttr[4] = {
{
// Add your model
...
3. And you will need to change the output of doInference in the common/algorithm/trt/trt_inference.cpp.
Please update the output_cov_buf and output_bbox_buf based on the YOLO architecture if needed.
And add the corresponding parser to generate the rectList.
You can find an example in our Deepstream sample:
/opt/nvidia/deepstream/deepstream-5.0/sources/objectDetector_Yolo/nvdsinfer_custom_impl_Yolo/nvdsparsebbox_Yolo.cpp
Thanks.
Hello, @AastaLLL
Okay, I’ll try it.
I have a question.
jetson multimedia api 04_video_dec_trt Can you provide a sample that uses onnx other than caffemodel in the sample?
Now
to video_dec_trt_main.cpp
include “NvOnnxParser.h”
…
using namespace nvonnxparser; I put in the build, but I will not be able to progress beyond that.
In the next article, I did what you gave me.
Can you provide a sample?
I have a question.
jetson multimedia api 04_video_dec_trt Can you provide a sample that uses onnx other than caffemodel in the sample?
Now
to video_dec_trt_main.cpp
include “NvOnnxParser.h”
…
using namespace nvonnxparser; I put in the build, but I will not be able to progress beyond that.
In the next article, I did what you gave me.
Can you provide a sample?
Thank you.
Hi,
You will need to update the algorithm/trt/trt_inference.cpp.
Please change the below function for the ONNX parser:
void
TRT_Context::caffeToTRTModel(const string& deployfile, const string& modelfile)
{
Int8EntropyCalibrator calibrator;
IInt8Calibrator* int8Calibrator = &calibrator;
// create API root class - must span the lifetime of the engine usage
IBuilder *builder = createInferBuilder(*pLogger);
INetworkDefinition *network = builder->createNetwork();
// parse the caffe model to populate the network, then set the outputs
ICaffeParser *parser = createCaffeParser();
...
The ONNX parser example can be found in the below sample:
/usr/src/tensorrt/samples/sampleOnnxMNIST/sampleOnnxMNIST.cpp
Thanks.
Hello,
It says “Converts the image from YUV to RGB format and saves it in a file”
The output of this sample program is known only as a txt file containing the bbox coordinate information.
Can I save it as an image?
I’m not sure what exactly the sentence underlined in red means.
Can you tell me the meaning of the sentence?
Thank you.
Hello, @AastaLLL
Hello.
04_video_dec_trt Can I save images other than the result*.txt file as explained in the example?
Thank you.
Hello,
https://docs.nvidia.com/jetson/l4t-multimedia/l4t_mm_vid_decode_trt.html
As the 32.5 release comes out, is it not necessary to proceed with this part?
Thank you.
Hello,
I guess I only understood the content of this answer.
I think I know the direction, but it doesn’t seem easy
Thank you.
Hi,
Sorry for the late update.
For 04_video_dec_trt, the output is text file only.
What kind of output do you want?
Do you want to dump the raw data image or the output with bbox marked?
The latter needs a nvosd component to draw the bbox over the frame.
You can find a demonstration in our backend example:
https://docs.nvidia.com/jetson/l4t-multimedia/nvvid_backend_group.html
Thanks.