Conversation
Was missing from the YOLO 2026 Hardware Acceleration section. Links to the dedicated yolo-detection-2026-coral-tpu skill.
- Add docker/Dockerfile: linux/amd64 image with edgetpu_compiler + ultralytics Converts yolo26n.pt → TFLite INT8 → yolo26n_int8_edgetpu.tflite - Add docker/compile.sh: one-shot runner that mounts models/ as output - Add docker/docker-compose.yml: alternative Compose-based runner - Add docker/README.md: full instructions for compile, commit, and use - Rewrite deploy.bat (Option A — no pycoral): * Remove vestigial install_pycoral.py step (detect.py uses ai-edge-litert directly) * Check for pre-compiled yolo26n_edgetpu.tflite from git repo * Falls back to SSD MobileNet download if YOLO model not yet compiled * Cleaner UAC handling with unique temp dir names to prevent collisions * Informational Python version check (not blocking — 3.9-3.13 all work) - Update requirements.txt: clarify no pycoral dependency, document libedgetpu - Update models/README.md: Docker compile workflow + expected output files
Iteration notes on edgetpu_compiler v16 compatibility: - TF 2.17 (onnx2tf >= 1.17): produces TFLite schema v6 → 'Invalid model' - TF 2.13 + onnx2tf: Keras symbolic-tensor TypeError during INT8 calibration - TF 2.13 + onnx-tf 1.10: strict=False conversion, SELECT_TF_OPS for YOLO ops - numpy randint dtype fix (uint8 → float32 cast for representative dataset) Docker image: python:3.9-slim + tensorflow-cpu==2.13.1 + onnx-tf==1.10.0 Pipeline: .pt → ONNX (opset 12) → SavedModel → TFLite INT8 → edgetpu_compiler Remaining issue: CONV_2D shape inference in onnx-tf SavedModel during calibration TODO: Use Google online compiler or EfficientDet-Lite as EdgeTPU-native backbone
…er pipeline
Model files (compiled via ultralytics format='edgetpu'):
models/yolo26n_full_integer_quant_edgetpu.tflite (2.7 MB, Edge TPU)
models/yolo26n_int8.tflite (3.7 MB, CPU fallback)
Docker / compile pipeline:
- Simplify Dockerfile: python:3.11-slim + ultralytics + edgetpu_compiler
Let ultralytics manage onnx2tf/TF versions (format='edgetpu' uses onnx2tf 1.28.8)
- Simplify compile_model.py: single call to model.export(format='edgetpu')
Per https://docs.ultralytics.com/guides/coral-edge-tpu-on-raspberry-pi/
- Removes all manual ONNX→SavedModel→TFLite pipeline complexity
detect.py:
- Updated _find_model_path to prefer *_full_integer_quant_edgetpu.tflite
(ultralytics naming convention) over legacy *_edgetpu.tflite pattern
Force-add to override .gitignore *.tflite exclusion. These pre-compiled models are committed to the repo so deploy.bat/deploy.sh can find them without needing a local Linux compilation environment. yolo26n_full_integer_quant_edgetpu.tflite (2.7 MB) — Edge TPU accelerated yolo26n_int8.tflite (3.7 MB) — CPU fallback Compiled with: ultralytics format='edgetpu', size=320, edgetpu_compiler v16
- Increase local /metrics scrape timeout to 5 minutes to accommodate server-side metric compilation for very large scale zero-copy graphs. - Add new cloud-provider-regression analysis skill.
Feature/coral tpu detection
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.