TI中文支持网Part Number:SK-TDA4VM
请问官方提供的post_process_image_object_detect.cpp里的void *PostprocessImageObjDetect::operator()(void *frameData, VecDlTensorPtr &results) 函数里如何得到目标检测模型输出的数据？是从results变量里解析得到吗？该如何解析呢？
Shine：
我把您的问题升级到英文e2e论坛了，请关注下面帖子的回复。https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1187417/tda4vm-void-postprocessimageobjdetect-operator-void-framedata-vecdltensorptr-results
,
Shine：
请看下面e2e工程师的回复。
I would recommend analyzing the function from the bottom up. Essentially, we are trying to get the following values from the results variable to overlay boxes on top of frameData:
numEntries – to determine how many objects are detected
score – to determine if we want to overlay box or ignore the detected object due to low confidence
label – to determine what class we are detecting
box – to determine top left and bottom right coordinates to know where to draw box
,
? ?：
我现在要把自己的后处理加进去，需要先获得推理的模型的输出层的数据即模型输出的特征，请问该怎么从results中提取出来呢？是这样吗？
auto *f_tensor = reinterpret_cast<float *>(results[0]->data);
f_tensor 是指向模型输出层数据的指针吗？如果是，指向的数据的类型和维度在这里是怎样的呢？
,
Shine：
已在e2e帖子上更新您的问题，请关注帖子的回复。
,
Shine：
抱歉回复晚了，请看下面e2e工程师的回复。
The logic that we have in our code to get the output is within the getVal function: https://www.ti2k.com/wp-content/uploads/ti2k/DeyiSupport_DSP_post_process_image_object_detect.cpp
I would recommend using similar, if not the same function to extract the output layer.
