• State Resolved
  • Locked Locked
  • Replies 4 replies
  • Answers 1 answer
  • Subscribers 99 subscribers
  • Views 666 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TDA4VM: Can i get pre-trained, not imported jpsdNet model?

Dasol Kim
Prodigy 165 points
Part Number: TDA4VM

Hi. 

I am testing AVP(Auto Valet Parking) app version 1 ~ 3  from vision_apps/dl_demos referring to here. I'm working on PSDKRA 7.0.0.11 release 

Q1. Can i get pre-trained the jpsdNet model? 

In case of AVP version1, jpsdNet and jsegNet each one works. Can i get the jpsdNet model ? (but not imported model by TIDL import tool)

If possible, I'll try to train jpsdNet model again. (Transfer learning or fine-tuning for my dataset.)

Q2. What is the difference between jDetNet and jpsdNet? 

If jpsdNet is not provided for public, can jDetNet model be re-trained for detecting Parking Slot? 

Are jpsdNet and jDetNet model completely different?

Q3. How about the onnx_tiad_ssd model of AVP v2~3 ? Can i get the onxx_tiad_ssd model?

What does 'tiad' mean? 

Best regards, 

Kim

  • 0
    Prodigy 165 points

    What I'm looking for is jpsdNet model.

    (I checked this link, the onnx model prefixed "tiad" means TI proprietary networks that are not available publicly.)

  • 0 in reply to Dasol Kim
    TI__Expert 4725 points

    Hello Kim,

    Please see my reply below specific to your questions.

    Q1. Can i get pre-trained the jpsdNet model? 

    >> we are planning to release the PyTorch training code for this in first quarter of next year. This is parking spot detection model for 512x512 input fish eye images. Model is based on MobileNetV1+ssd with addition of extra four channels (in location head) for four corners of parking spot. More details about this nw can be found at

    https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=9012933

    Q2. What is the difference between jDetNet and jpsdNet? 

    >>jDetNet is generic ssd based object detection model in caffe framework. where as jpsdNet is just parking spot detection model in PyTorch.

    Q3. How about the onnx_tiad_ssd model of AVP v2~3 ? Can i get the onxx_tiad_ssd model?

    >> onnx_tiad_ssd is similar model as jpsdNet. It can detect vehicles along with parking spot detection. This is targeted for AVP demo. Resolution here is 768x384 to match with other models( semantic segmentation) of AVP demo. Going forward we recommend to use this model instead of jpsdNet, as it can detect parking spots. Proposed released framework in first quarter of 2021, will support model training for jpsdNet and onnx_tiad_ssd both.

    >> tiad is internal TI prefix naming convention for AVP demo models.

    Regards

    Deepak Poddar

  • 0 in reply to Deepak Poddar
    Prodigy 165 points

    Hello, Poddar.

    Thanks for the support. It was helpful.

  • 0 in reply to Dasol Kim
    TI__Expert 4725 points

    Hi,

    we have released the training code with pretrained weights at

    https://git.ti.com/cgit/jacinto-ai/pytorch-ssds-keypoints/about/

    Regards

    Deepak Poddar