ARG-NCTU/fcl_rtdetr_r10_e100_480

FCL RT-DETR

Model Information

• Model: best.pt
• Dataset: CL0: SMD / CL1: SPSCD
• #Classes: 6(SPSCD) + 6(Shared) + 4(SMD)
• Image Size: 480
• Total Training Epochs: 100
• Total Rounds: 10

Training Time

• CL0: SMD
  • Hardware: NVIDIA RTX 5070 12GB
  • Total Duration: 401h 4m 46s (1,443,887s)
  • Average per Epoch: 240.6 minutes
• CL1: SPSCD
  • Hardware: NVIDIA RTX 5090 32GB
  • Total Duration: 205h 30m 6s (739,806s)
  • Average per Epoch: 123.3 minutes

Global Validation Results

- last_rnd_global/:
    - BoxF1_curve.png
    - BoxPR_curve.png
    - confusion_matrix_normalized.png
    - confusion_matrix.png

Repository Structure

• c0_smd/: Result for SMD training stage.
• c1_spscd/: Result for SPSCD training stage.
• last_rnd_global/: Global validation metrics & curves (F1, PR, Confusion Matrix).
• weights/: Contains best.pt and last.pt.

How to Use

from ultralytics import RTDETR

# Load model
model = RTDETR('best.pt')

# Run inference
results = model('your_image.jpg')