- calibrate_probs_selects_top_label: use logits that produce confidence >= 0.70
- prediction_response: forward status from Prediction struct instead of hardcode
Co-Authored-By: Claude <noreply@anthropic.com>
- PredictionResponse::status was missing in prediction_response()
- Prefix preprocess_start with underscore for unused variable warning
Co-Authored-By: Claude <noreply@anthropic.com>
- Integrate Convolutional Block Attention Module (CBAM) for improved feature focus
- Implement temperature scaling and confidence-based status reporting
- Automate dataset acquisition using kagglehub
- Update ONNX opset to 18 and refine preprocessing validation
- Store ONNX Session in Option<Mutex<Session>> for thread-safe concurrent access
- Lock session before run() in predict(), map lock poisoning to PredictionFailed
- Preserve ModelUnavailable when session is None
- Reject non-finite values (NaN, +inf, -inf) in prediction_from_probabilities
- Add unit tests for NaN, positive infinity, and negative infinity rejection
- Keep all existing tests passing
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Create Rust crate skeleton with Cargo.toml, main.rs, and config.rs.
Includes package metadata, dependencies (anyhow, axum, image, ndarray,
ort, serde, tokio, tower, tracing), and config module with LABELS
constants, SERVICE_NAME, SERVICE_VERSION, DEFAULT_INPUT_SIZE, and
resolve_model_path function. Config tests verify label order, relative
path resolution, and absolute path preservation.
Also includes approved spec and implementation plan documents.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
