How to use pretrained model i trained on my cutom dataset

[VyrionAI1]

Search before asking

• I have searched the RF-DETR issues and found no similar feature requests.

Description

I have successfully trained my custom model. After making some changes to my dataset, I am now starting a fresh training session using my pre-trained model as a starting point. This is a new training run, not a resumed one.

Use case

No response

Additional

No response

Are you willing to submit a PR?

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

[VyrionAI1]

i uss this even that is start from scratch

from rfdetr import RFDETRSegPreview

model = RFDETRSegPreview()

model.train(dataset_dir="/content/-X_MERGE+caninedataset_june_2025-5",resolution=432 , epochs=50, batch_size=4, grad_accum_steps=2,run_test=False,output_dir='/content/drive/MyDrive/TRANING/RF_DETR/output432',pretrain_weights="/content/drive/MyDrive/TRANING/RF_DETR/output/checkpoint_best_total.pth")

[Abdul-Mukit]

I believe you need to set pretrain_weights argument when you initialize the model instead of passing it to train.
model = RFDETRSegPreview(pretrain_weights=xyz)

The overall organization of parameters in RF-DETR is confusing. There is one set of configurations in model config, another in training config, and then there are some in the argument parser. Sometimes they overlap.