The slow progress of RF-DETR training and the post-training validation process

Search before asking

• I have searched the RF-DETR issues and found no similar bug report.

Bug

I have two questions: 1) I ran a training session using RF-DETR with my own custom dataset. Even though I ran the training on a GPU, it took a very long time. My system has a “Tesla V100-PCIE-16GB” graphics card.

The code I ran for training:

from rfdetr import RFDETRSegLarge

model = RFDETRSegLarge()

model.train(
    dataset_dir="Data",
    output_dir="runs/rfdetr_seg_large",
    epochs=50,
    batch_size=2,
    grad_accum_steps=8,
    lr=1e-4
)

After training the RF-DETR model and obtaining “checkpoint_best_total.pth,” is there an easy way to perform validation, just like in YOLO training? How can I perform validation on the test data using the model I obtained after training?

Environment

RF-DETR: 1.8.1

OS: Ubuntu 22.04.4 LTS

Python: 3.10.12

PyTorch: 2.10.0+cu126

GPU: NVIDIA Tesla V100-PCIE-16GB

Minimal Reproducible Example

from rfdetr import RFDETRSegLarge

model = RFDETRSegLarge()

model.train(
    dataset_dir="Data",
    output_dir="runs/rfdetr_seg_large",
    epochs=50,
    batch_size=2,
    grad_accum_steps=8,
    lr=1e-4
)

Additional

No response

Are you willing to submit a PR?

• Yes, I'd like to help by submitting a PR!