TDA4VM: [edgeai-modelmaker Compile Issue] Invalid Layer Name Error on TiSCAPES2017_driving Model (0% Accuracy)

wang haofei

Part Number: TDA4VM

Tool/software:

Dear Community,

I'm encountering an issue while trying to train the TiSCAPES2017_driving dataset using the edgeai-modelmaker toolkit. Here are my steps and the error:

Environment: edgeai-tensorlab/edgeai-modelmaker/
Command: ./run_modelmaker.sh TDA4VM config_detection.yaml
Result: Model compiles but shows 0% accuracy with the error:
```
Invalid Layer Name: /multi_level_conv_obj.2/Conv_output_0
```

Attached is my complete run.log for reference.

Has anyone encountered similar layer naming issues during compilation? Any suggestions for resolving this accuracy problem would be greatly appreciated.

Thank you in advance for your help!
Best regards,
HaoFei W

5 months ago

0 wang haofei 5 months ago

Prodigy 50 points

1727.run.log

0 Chris Tsongas 5 months ago in reply to wang haofei

TI__Genius 14660 points

Can you include your ONNX omdel?

Regards,

Chris

0 wang haofei 5 months ago in reply to Chris Tsongas

Prodigy 50 points

od-8200.tar.gz
Hi Chris, here's my ONNX model.

0 wang haofei 4 months ago in reply to Chris Tsongas

Prodigy 50 points

Hello Chris, my problem hasn't been solved yet. Could you help me?

0 Christina Kuruvilla 4 months ago in reply to wang haofei

TI__Expert 5440 points

Hi Wang, due to Chris' bandwidth, I will be assisting. Is this being run on PC emulation on edgeai-benchmark? Have you tried it with the latest version? Could you also share the .yaml as well?

Warm regards,

Christina

0 wang haofei 4 months ago in reply to Christina Kuruvilla

Prodigy 50 points

Yes, I ran this on a PC using edgeai-benchmark emulation. I’ve tested with the latest version, but the issue persists, and the accuracy remains 0.

Attached is my .yaml configuration file for reference.

common:
target_module: 'vision'
task_type: 'detection'
target_device: 'TDA4VM'
# run_name can be any string, but there are some special cases:
# {date-time} will be replaced with datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
# {model_name} will be replaced with the name of the model
run_name: '{date-time}/{model_name}'

dataset:
# enable/disable dataset loading
enable: True #False
# max_num_files: [750, 250] #None

# Object Detection Dataset Examples:
# -------------------------------------
# Example 1, (known datasets): 'widerface_detection', 'pascal_voc0712', 'coco_detection', 'udacity_selfdriving', 'tomato_detection', 'tiscapes2017_driving'
# dataset_name: widerface_detection
# -------------------------------------
# Example 2, give a dataset name and input_data_path.
# input_data_path could be a path to zip file, tar file, folder OR http, https link to zip or tar files
# for input_data_path these are provided with this repository as examples:
# 'software-dl.ti.com/.../tiscapes2017_driving.zip'
# 'software-dl.ti.com/.../animal_detection.zip'
# -------------------------------------
# Example 3, give image folders with annotation files (require list with values for both train and val splits)
# dataset_name: coco_detection
# input_data_path: ["./data/projects/coco_detection/dataset/train2017",
# "./data/projects/coco_detection/dataset/val2017"]
# input_annotation_path: ["./data/projects/coco_detection/dataset/annotations/instances_train2017.json",
# "./data/projects/coco_detection/dataset/annotations/instances_val2017.json"]
# -------------------------------------
dataset_name: tiscapes2017_driving
input_data_path: 'software-dl.ti.com/.../tiscapes2017_driving.zip'
#dataset_name: broken
#input_data_path: "./data/projects/broken/dataset"

training:
# enable/disable training
enable: True #False

# Object Detection model chosen can be changed here if needed
# options are: 'yolox_s_lite', 'yolox_tiny_lite', 'yolox_nano_lite', 'yolox_pico_lite', 'yolox_femto_lite', 'yolov7_l_lite'
model_name: 'yolox_nano_lite'

training_epochs: 15 #30
# batch_size: 8 #32
# learning_rate: 0.005
num_gpus: 1 #1 #4

compilation:
# enable/disable compilation
enable: True #False
# tensor_bits: 8 #16 #32

0 Christina Kuruvilla 4 months ago in reply to wang haofei

TI__Expert 5440 points

Thank you for this. I will try to recreate your issue on my end.

Warm regards,

Christina

0 Rohit Rao 3 months ago in reply to wang haofei

TI__Prodigy 210 points

Hi Wang,

We will try to resolve your other E2E request first, which may resolve this as well.

e2e.ti.com/.../tda4vm-compilation-failure-in-edgeai-modelmaker-10-1r-cpu

Best,

Rohit

0 Akash Dhande 2 months ago in reply to Rohit Rao

Prodigy 20 points

Hello any update on this issue? I am facing a similar issue in which I have trained my YOLOX-Nano model using edgeai-mmdetection, exported the ONNX and prototxt, and lastly used the tutorials from the edgeai-benchmark to compile it. After compiling the model without any quantization i.e at 32 bits, it still gives me 0 accuracy. However before compilation, the training accuracy is above 50%. This seems like a critical issue to me making the entire stack almost unusable to get a working model. Please suggest a solution or alternatives.

0 Christina Kuruvilla 2 months ago in reply to Akash Dhande

TI__Expert 5440 points

Hi Akash,

The modelmaker honestly has a few bugs that need to be looked at but is in queue to be investigated. I will create a Jira ticket for Dev team to look through but this may take some time. My colleague Rohit created an alternate script to use on this E2e in the meantime: e2e.ti.com/.../tda4vm-compilation-failure-in-edgeai-modelmaker-10-1r-cpu

Warm regards,

Christina

Processors

Processors forum

TDA4VM: [edgeai-modelmaker Compile Issue] Invalid Layer Name Error on TiSCAPES2017_driving Model (0% Accuracy)