TDA2EXEVM: How can I train the dataset of traffic sign with objSubType？

Focus_lee

Part Number: TDA2EXEVM

Hi,

In TIDL_OD demo, the target recognition is classified as follows:

typedef enum
{
    TI_OD_PEDESTRIAN = 0,
    TI_OD_TRAFFIC_SIGN,
    TI_OD_VEHICLE,
    TI_OD_MAX_OBJECTS
} TI_OD_ObjectType;

I am now have a car and traffic sign(only speed limit sign 30,40,50,60,80,100) recognition task, and My classification is as follows：

But there are objSubType in the property list as follows：

How do I attribute my speed limit sign classification to TSR?

typedef struct
{
uint32_t objTag;
uint32_t objType;
uint32_t objSubType;
uint32_t xPos;
uint32_t yPos;
uint32_t objWidth;
uint32_t objHeight;
float    objScore;
float    objScale;
float    reserved0;
float    reserved1;
float    reserved2;
} TI_OD_objectDescriptor;

over 4 years ago

0 Focus_lee over 4 years ago

Expert 1055 points

I am now use the ssdJacintoNetV2 net in caffe-jacinto-model.

0 Manu Mathew over 4 years ago in reply to Focus_lee

TI__Genius 9926 points

Is this question about training? Or is it about the actual inference in TIDL? Can you please describe your question more clearly.

Thanks & regards,

0 Focus_lee over 4 years ago in reply to Manu Mathew

Expert 1055 points

yes, My question both about training and inference in tidl .

That TIDL_OD example is the identification of pedestrian, vehicles and traffic signs. We all know that there are many different types of traffic signs，So TI divided traffic signs into a subclass to determine which kind of signs.

I just want to know how can I use the structure provided by TI to set my classification, and the result of output also includes the result of subclass。

0 Praveen Eppa1 over 4 years ago in reply to Focus_lee

TI__Genius 17580 points

I would suggest to have different classes for different speed limit signs to use TIDL.

Thanks,

Praveen

0 Focus_lee over 4 years ago in reply to Praveen Eppa1

Expert 1055 points

Hi,

There's still a problem，In the identification of traffic signs, one kind of images often make mistakes, resulting in low identification errors.

just show as blow. The left image is belong to the speed limit signs, which I need to recognize in my task, but the right image is very similar to the left，which not in my task. How can I strengthen my training for this situation？The right image is not marked in my sample.

Can I use subclass attributes to reinforce the training of the sample?

0 Focus_lee over 4 years ago in reply to Praveen Eppa1

Expert 1055 points

And, We want to get a good result of vehicle identification at night，For ssdJacintoNetV2 networks, I need to separate night vehicles into another one category or a subcategory，or just in the car category without subdivision？