The image classification model deployed on the ESP32 will report an error after running for the third time in the while loop

Help
#1

Question/Issue:
[Describe the question or issue in detail]
The image classification model deployed on the ESP32 will report an error when run for the third time
Project ID:
[Provide the project ID]
827169
Context/Use case:
[Provide context or use case where the issue is encountered]
When deploying a sample project using ESP32, I can enter the main loop, but an error occurs during the third run
Steps Taken:

  1. [Step 1]
    Complete the training of my own classification task in Edge Impulse and generate a C++ link library
  2. [Step 2]
    Copy the generated library to the project at git clone GitHub - edgeimpulse/example-standalone-inferencing-espressif-esp32: Builds and runs an exported impulse locally (ESP IDF)
  3. [Step 3]
    Modify the kTensorArenaSizein
    4.[Step 4]
    When the program is burned into the ESP32S3 for operation, the following error occurs
    ๅ›พ็‰‡
    ๅ›พ็‰‡730ร—429 71.1 KB

Environment:

  • **Build Environment Details: ESP IDF 5.4.2
  • **OS Version:Ubuntu 20.04
  • Edge Impulse Version (Firmware): โ€‹latest version**
#2

This is all the error messages

ๅ›พ็‰‡
ๅ›พ็‰‡1836ร—312 186 KB

#3

ๅ›พ็‰‡
ๅ›พ็‰‡1820ร—846 257 KB

#4

ๅ›พ็‰‡
ๅ›พ็‰‡777ร—100 29.2 KB

When I was looking at the document, I noticed that the feature in the document is a floating-point decimal, but mine is in hexadecimal. Could this be the issue

#5

Hello @nova1231 first of all welcome to the Edge Impulse community!

Thanks for reporting the detailed issue and then the potential issue.

Let us know if that can fix your issue!

#6

first of all,appreciate your reply .today i rebuild a new project of audio detect not image , it`s working correctly.here is the image of working.

ๅ›พ็‰‡
ๅ›พ็‰‡831ร—846 152 KB

#7

i feel confused and want run image task finally.do you have any idea about this?

#8

Hi, @nova1231 !
Floating-point decimal is correct type for images (itโ€™s three bytes - R, G, B values - crammed into a float), no problem there.
While Iโ€™m trying to reproduce the error, one thing to try is to change class names from Chinese to English. We donโ€™t really test for anything except labels in English language.

1 Like
#10

i have tested English label,but still meeting same question.as the image show

ๅ›พ็‰‡
ๅ›พ็‰‡721ร—367 90.6 KB

The difference is that this time it reported an error upon entering

#11

@nova1231
I cannot access your project, it looks like it was deleted.

#12

sorry i deleted the project unguarded. 832152 is the new image proect ID,still have same question.

#13

Okay, so Iโ€™m doing the testing using fresh installation of 5.1 ESP-IDF.
You can see 5.1 requirement here

version above might work, but not guaranteed, as not tested.

I found an issue with calloc/alloc, that was due Epressif adding more changes to release/5.1 branch.
Iโ€™m making PR to fix this, but meanwhile you can
change
#if ESP_IDF_VERSION >= ESP_IDF_VERSION_VAL(5, 2, 1)
to
#if ESP_IDF_VERSION >= ESP_IDF_VERSION_VAL(5, 0, 0)
in
edge-impulse-sdk/porting/espressif/ei_classifier_porting.cpp of your project.

To summarize:

  1. Use ESP IDF 5.1
  2. Make the above change. Later the required change will be merged into our SDK and manual change will be not needed.

Let me know of it works!

#14

i changed #if ESP_IDF_VERSION >= ESP_IDF_VERSION_VAL(5, 2, 1) to #if ESP_IDF_VERSION >= ESP_IDF_VERSION_VAL(5, 0, 0) and using idf 5.1.6 version. now ,it`s working.

ๅ›พ็‰‡
ๅ›พ็‰‡1543ร—469 69.1 KB

ๅ›พ็‰‡
ๅ›พ็‰‡1037ร—40 10.2 KB

#16

If this issue is resolved in idf5.4 version, please let me knowใ€‚