23 days ago
-
Add CoreML category classifier (category_clf_b8_s128.mlpackage)
-
Add CoreML category classifier (category_clf_b8_s64.mlpackage)
-
Add CoreML category classifier (category_clf_b8_s32.mlpackage)
-
Remove mis-nested ppocr_rec_v6_cal.mlpackage (downloader expects it at repo root)
-
Add CoreML mlpackage for PP-OCRv5 server rec (ppocr_rec_v6_cal.mlpackage)
-
Add CoreML mlpackage for PP-OCRv5 server rec (ppocr_rec_v6_cal.mlpackage)
-
Add MobileCLIP S2 ONNX (LICENSE) for Windows auto-download (from Xenova/mobileclip_s2, Apple AMLR license included)
-
Add MobileCLIP S2 ONNX (tokenizer.json) for Windows auto-download (from Xenova/mobileclip_s2, Apple AMLR license included)
-
Add MobileCLIP S2 ONNX (text_model.onnx) for Windows auto-download (from Xenova/mobileclip_s2, Apple AMLR license included)
-
Add MobileCLIP S2 ONNX (vision_model.onnx) for Windows auto-download (from Xenova/mobileclip_s2, Apple AMLR license included)
-
text_detector: yolo26l-obb retrained on ds6 (california, 9,663 pages, 1.68M ghost-purged boxes). Gold val P .982 R .826 mAP50 .833 mAP50-95 .808; coverage recall @50% = 99.03%. Export: imgsz 1536, max_det=1000, end2end, output (1,1000,7). Prior version: lfs a340c0259dde082e (rollback via revision 9e300531484e).
-
Upload ppocr_v6_cal/ppocrv6_dict.txt with huggingface_hub
-
Upload ppocr_v6_cal/ppocr_fisher_v6_cal.onnx with huggingface_hub
-
Add finetuned PP-OCRv6_medium_rec (number/symbol-augmented, fp32+LR-drop): general 0.986, numbers 95.1% (beats v5 0.935/93.1% on every pattern)
-
Add finetuned PP-OCRv6_medium_rec (number/symbol-augmented, fp32+LR-drop): general 0.986, numbers 95.1% (beats v5 0.935/93.1% on every pattern)
-
Remove stale native CoreML .mlpackage (rec served as fine-tuned ONNX)
-
Deploy fine-tuned PP-OCRv5 server rec model (replaces stale)
-
Add ANE-accelerated CoreML LayoutLMv3 (fp32 embed+bias / fp16 ANE encoder, 3 seq buckets)
-
Replace item_detector with YOLO26l-obb @1280 (4-class: sale_date/sale_item_info/promotions/category)
-
Add finetuned PP-OCRv5 rec (number/symbol-augmented, balanced fp32): general 0.935, numbers 93.1% (base 71%), cents 13->98%
-
Add CoreML mlpackage for PP-OCRv5 server rec (fp32, GPU-accelerated; ~15x faster than ONNX-CPU, output-identical)
-
Update text_detector.onnx: OBB model with max_det=1000 (removes 300-detection cap)
-
OCR post-process dictionary: EN+ES+domain (1.01M words)
-
ppocr_rec_v5_server: epoch-20 final (val acc 0.959, norm_edit 0.994)
-
Replace text_detector with yolo26 OBB (end2end, 1536)
-
Upload ppocr_rec_v5_server/ppocrv5_dict.txt with huggingface_hub
-
Upload ppocr_rec_v5_server/model.onnx with huggingface_hub
-
Upload ppocr_v5/ppocr_fisher.onnx
-
Delete ppocr_rec
purging old ocr
-
Upload 3 files
-
Delete birefnet_lite_exact_fp16_512.mlpackage
-
Upload 3 files
-
Upload 3 files
-
Upload BiRefNet_lite-general-2K-epoch_232.onnx
-
Delete BiRefNet_HR_1024.onnx.data
-
Delete BiRefNet_HR_1024.onnx
-
Upload BiRefNet_HR_1024.onnx.data
-
Upload BiRefNet_HR_1024.onnx
-
Upload item_detector.onnx
-
Upload ppocr_rec/model_simplified.onnx
-
Upload 8 files
-
Upload 16 files
-
Delete final_model
-
Delete roberta_description
-
Delete roberta_catergory
-
Upload 4 files
-
Upload 7 files
-
Upload item_detector.onnx
-
Upload item_detector.onnx
-
Upload item_detector_medium.onnx
