docs: bird-classifier card lists the Edge TPU build

animalClassification/bird-classifier/README.md

@@ -10,7 +10,23 @@
 | Code licence | Apache-2.0 |
 | **Weights licence** | Apache-2.0 ([`LICENSES/Apache-2.0.txt`](../../LICENSES/Apache-2.0.txt)) |
 | Training data | not stated upstream |
-| CamStack modifications | converted from TensorFlow with tf2onnx; OpenVINO IR and CoreML derived from the ONNX |
+| CamStack modifications | converted from TensorFlow with tf2onnx; OpenVINO IR and CoreML derived from the ONNX; the Edge TPU build (`tflite/`) is the Google full-integer quantised TFLite of the same model, compiled with `edgetpu_compiler` 14 |
 | Commercial use | permitted by Apache-2.0 |
 
+## Formats
+
+| Format | File | Notes |
+| --- | --- | --- |
+| ONNX | `onnx/bird-classifier.onnx` (+ `bird-classifier-labels.json`) | fp32, NHWC, input `x/255` |
+| OpenVINO IR | `openvino/bird-classifier.{xml,bin}` | from the ONNX |
+| CoreML | `coreml/bird-classifier.mlpackage` | from the ONNX |
+| Coral Edge TPU | `tflite/bird-classifier_edgetpu.tflite` | full-integer uint8; input raw 0..255 RGB (scale 1/128, zero point 128 in the graph); output `[1,965]` uint8 softmax, scale 1/256 — dequantise, do not re-softmax. All 65 operations on the Edge TPU, 3.75 MiB of parameter cache |
+
+The Edge TPU file is built from `mobilenet_v2_1.0_224_inat_bird_quant.tflite`
+(google-coral/test_data commit `104342d2`, sha256 `350fcd8c…04f9`) by
+`scripts/build-bird-classifier-edgetpu.py` in the CamStack repository. Every
+compile writes a fresh random parameter-cache token, so a rebuild differs from
+this file in those 16 bytes only. Labels: the ONNX folder's labels file (same
+965 entries, same order).
+
 CamStack redistributes the files in this folder; every change from the upstream is stated above. Converting a model does not change its licence. The collection statement is [`LICENSE`](../../LICENSE); every model's attribution is in [`THIRD_PARTY_MODELS.md`](../../THIRD_PARTY_MODELS.md).