> ESP32-S31 is particularly well suited for edge AI and machine learning workloads, including neural network inference

Any way to know what kind of performance one could expect running e.g. a depth anything model on there?

A real example https://github.com/OHF-Voice/micro-wake-word

Regarding specifically depth anything: You're not running this on a microcontroller. In general, CNNs still reign supreme on microcontrollers since you have a way lower peak memory demand which is what usually kills you. Here in this case you have a couple of _kilobytes_ of SRAM, potentially extendable to a couple of megabytes of PSRAM.

Even for small CNNs you often need to do some quite complex interleaving of layers (i.e. running parts of layer 1 and layer 2 in parallel interleaved to take advantage of the downsampling of CNNs) to keep performance and memory impact reasonable (see e.g. https://openreview.net/pdf?id=2O8qbyxH6X).

Think more "image classifier" less "run an image to image transformer". For depth anything, a single layer's activation is probably significantly larger than the available SRAM (I think it is (224/16)^2 patches each with activations [48, 96, 192, 384] for depth anything small: You aren't running this.)

I was wondering this as well. What exactly makes this a good AI chip vs others.

Unless they're not listing a major feature in their spec, a dual core 320Mhz microcontroller is not bad but youre not going to be running any kind of vision model on it, at least very fast.

Memory is the main constraint. You have what, 8mb of psram.

Compute wise you can manage. You can do quantisation and run a small 10-15 layer CNN perhaps. Image classification is possible. Keep in mind the channel count and input resolution cannot be high since memory will be a problem. You can maybe do face _detection_, "is my cat on my keyboard" classification as well maybe.

Audio, you can do a lot more. Wake word detection happens on _much_ smaller accelerators inside iphones. In this one you can do slightly heavier classifications. Maybe speaker identification "which member of family" or maybe "which dog is barking"

nope. not happening. at most YOLO or mayyybe FastDepth