Update validated TorchVision exports and compatibility documentation

Publish the reviewed model files and current CPU/GPU/NPU compatibility table.

README.md

@@ -43,9 +43,9 @@ The ResNet-50 architecture is a convolutional neural network pre-trained on the
 
 The model was converted from a checkpoint from PyTorch Vision.
 
-The original model has:
-acc@1 (on ImageNet-1K): 76.13%
-acc@5 (on ImageNet-1K): 92.862%
+The source checkpoint has:
+acc@1 (on ImageNet-1K): 80.858%
+acc@5 (on ImageNet-1K): 95.434%
 num_params: 25,557,032
 
 ## Available Model Files
@@ -74,14 +74,12 @@ Calibration used real ImageNet validation images with the TorchVision ResNet
 preprocessing flow. When using APIs that expose raw tensor buffers, prepare the
 input and output using the quantization parameters stored in the model.
 
-## Runtime Compatibility
+## Compatibility
 
-These artifacts are intended for LiteRT CPU and GPU execution. The static INT8
-channelwise artifact is also suitable for Qualcomm NPU deployment through the
-LiteRT Qualcomm compiler plugin and QNN AOT compilation on compatible Qualcomm
-devices.
-
-Enablement for other NPU backends is still under validation.
+| File | CPU | GPU | NPU |
+|---|---|---|---|
+| `resnet50.tflite` | Supported | Supported | N/A |
+| `resnet50_int8_channelwise.tflite` | Supported | Not supported | Qualcomm / MediaTek |
 
 ## Intended uses & limitations
 
@@ -111,20 +109,20 @@ from ai_edge_litert.compiled_model import CompiledModel
 def preprocess(img: Image.Image) -> np.ndarray:
    img = img.convert("RGB")
    w, h = img.size
-   s = 256
+   s = 232
    if w < h:
-       img = img.resize((s, int(round(h * s / w))), Image.BILINEAR)
+       img = img.resize((s, int(h * s / w)), Image.BILINEAR)
    else:
-       img = img.resize((int(round(w * s / h)), s), Image.BILINEAR)
-   left = (img.size[0] - 224) // 2
-   top = (img.size[1] - 224) // 2
+       img = img.resize((int(w * s / h), s), Image.BILINEAR)
+   left = int(round((img.size[0] - 224) / 2.0))
+   top = int(round((img.size[1] - 224) / 2.0))
    img = img.crop((left, top, left + 224, top + 224))
 
    x = np.asarray(img, dtype=np.float32) / 255.0
    x = (x - np.array([0.485, 0.456, 0.406], dtype=np.float32)) / np.array(
        [0.229, 0.224, 0.225], dtype=np.float32
    )
-   return np.transpose(x, (2, 0, 1))
+   return np.ascontiguousarray(x.transpose(2, 0, 1)[None])
 
 def main():
    ap = argparse.ArgumentParser()

resnet50.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12d7455c0db04121af403c69cf3df90254536e79cbc0ea81eaa8d80ab1f42d68
-size 102148096
+oid sha256:b60ece2473934776b8de38244d1fc5d107ecebabd7080a1a19ea8c28a4643be5
+size 102147184

resnet50_int8_channelwise.tflite

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b47a9100b1ccd66c1dd5ad59c2ef6bc1ae43dbda240cd9354e1ea874a6f0ebc
-size 26307408
+oid sha256:b5a5bcd4cc610c8bc8afb48f8d220a1716a4a0d752b6d9f3ab0e7a8e65d4f23a
+size 26305808