Student: 김한별 (202413050). Base: facebook/detr-resnet-50. Dataset revision: 5c0e1d3d0f5a95bb916995c8ad34778cd2ca6c7c. Profile: final; epochs: 100. Training/validation/test kept separate. Best epoch selected by validation loss. Measured test metrics: {"metrics": {"AP": 0.8988351952057015, "AP50": 0.9868176484675787, "AP75": 0.9601280229832913, "AP_small": -1.0, "AP_medium": -1.0, "AP_large": 0.8988351952057015, "AR1": 0.9278985507246377, "AR10": 0.9304347826086957, "AR100": 0.9304347826086957, "AR_small": -1.0, "AR_medium": -1.0, "AR_large": 0.9304347826086957}, "per_class_AP": {"furniture": null, "Chair": null, "Sofa": 0.9137572724663771, "Table": 0.8839131179450259}} Parameter counts and runtime: {"total_params": 41502409, "trainable_params": 18047497, "trainable_ratio": 0.43485420328251306, "session_seconds": 3374.337096424, "resumed": false, "peak_vram_gb": 0.885479424, "best_validation_loss": 0.1884491890668869, "best_checkpoint": "/content/drive/MyDrive/202413050_detr/final/partial/checkpoints/checkpoint-4104", "epoch": 100.0, "train_metrics": {"train_runtime": 3373.8575, "train_samples_per_second": 13.456, "train_steps_per_second": 1.689, "total_flos": 7.80937305172992e+18, "train_loss": 0.2154312148847078, "epoch": 100.0}, "environment": {"gpu": "Tesla T4", "torch": "2.11.0+cu130", "profile": "final", "epochs": 100, "image_size": 480, "effective_batch_size": 8, "seed": 42}} Dataset licensing: consult its original Roboflow page; dataset card lists cc. This is an educational detector; performance may differ on cluttered real rooms.
model.safetensors 166 MBtraining_args.bin 5.9 kB