ResNet-50 (ANEForge)

An unmodified duplicate of microsoft/resnet-50, tagged for use with ANEForge so the weights load and run directly on the Apple Neural Engine (no CoreML). BatchNorm is folded into the preceding convolution at load. Weights are byte-identical to the source; see the original repo for the model details. Licensed apache-2.0 from the source.

Use with ANEForge

# pip install "aneforge[models]"
import aneforge as af
from transformers import AutoImageProcessor

proc = AutoImageProcessor.from_pretrained("aneforge/resnet-50")
net = af.load_resnet("aneforge/resnet-50")
logits = net(proc(image, return_tensors="np")["pixel_values"])   # image: a PIL.Image -> [1, 1000] logits

ANEForge compiles the model's graph into a single ANE program and streams the weights from this repo via huggingface_hub. See the docs and the paper.