Sorry for the late reply.
More or less. There may be insignificant differences between Studio and on the ESP32-CAM on how we crop/scale the image, but not enough to cause incorrect classification results.
I’d first start of with our example standalone for esp32. Copy raw features from your project and paste them in the features array. And confirm you get the same results as Studio.
Then work your way up with the actual sensor trying to genereate features to feed the classifier. Again, Louis’ project is a good working example