去背積木 Computer Vision Model
Dataset Version
Generated on May 5, 2026
Notes: 依照GTP協助增加對黑色積木有效優化的前處理與資料增強選項 {核心問題：類別內的「變異性」太大 (Intra-class Variance) 目前的狀況是：模型試圖在一個標籤下，同時學習「會閃光的平滑方塊」和「不閃光的粗糙破掉方塊」。 結果： 模型找不到一個「公約數」特徵。在光線亮時，樂高的反光太強，3D 列印的又太黑，模型最終決定「保守起見」，信心值達不到標準，所以選擇不框選（漏檢）。
Popular Download Formats
YOLO26
TXT annotations and YAML config used with YOLO26.
YOLOv12
TXT annotations and YAML config used with YOLOv12.
YOLOv11
TXT annotations and YAML config used with YOLOv11.
YOLOv9
TXT annotations and YAML config used with YOLOv9.
YOLOv8
TXT annotations and YAML config used with YOLOv8.
YOLOv5
TXT annotations and YAML config used with YOLOv5.
YOLOv7
TXT annotations and YAML config used with YOLOv7.
COCO JSON
COCO JSON annotations are used with EfficientDet Pytorch and Detectron 2.
YOLO Darknet
Darknet TXT annotations used with YOLO Darknet (both v3 and v4) and YOLOv3 PyTorch.
Pascal VOC XML
Common XML annotation format for local data munging (pioneered by ImageNet).
TFRecord
TFRecord binary format used for both Tensorflow 1.5 and Tensorflow 2.0 Object Detection models.
PaliGemma
PaliGemma JSONL format used for fine-tuning PaliGemma, Google's open multimodal vision model.
CreateML JSON
CreateML JSON format is used with Apple's CreateML and Turi Create tools.
Other Formats
Choose another format.
1317 Total Images
View All Images
Dataset Split
Train Set 88%
1155Images
Valid Set 8%
107Images
Test Set 4%
55Images
Preprocessing
Auto-Orient: Applied
Resize: Stretch to 640x640
Grayscale: Applied
Auto-Adjust Contrast: Using Adaptive Equalization
Random Sample: Include 100% train, 100% valid, 100% test
Augmentations
Outputs per training example: 3
Rotation: Between -45° and +45°
Grayscale: Apply to 20% of images
Brightness: Between -25% and +25%
Exposure: Between -25% and +25%
Blur: Up to 2.5px
Noise: Up to 5% of pixels