prashant-dn

> Please provide me with some reference code on how to use the pretrained checkpoint model Have you gone through this part? 

@AnishMathewOommen Why do you want to train the model on obj365 data? The pre-trained weights suffice there. You are probably looking for fine-tuning these models on your custom dataset or...

It is specified in the README but here is a snippet: Once you have downloaded the pre-trained weight of your choice from the links mentioned in the README, ```bash i=3_906_52_2024-12-08-17-25-14_1800_OG_0148.jpg...

Same issue +1

I solved it this way. 1. My train json was truncated... had to re-generate it. 2. The issue says that some tensors (in gpu) can't be accessed hence the assertion...

@nikky4D first ensure that your annotation json is a valid json. Second run the python3 command using the flags `TORCH_USE_CUDA_DSA=1` and `CUDA_LAUNCH_BLOCKING=1` as shown in above comment.

@nikky4D Actually the problem is the labels - The class ids in my COCO annotation json file were not 0-indexed i.e., the first class was 1 and not 0, while...

Same here - I tried these 2 variants : `dfine_hgnetv2_${model}_obj365.yml` and `dfine_hgnetv2_${model}_custom.yml` and the mAP50 never goes above 0.35 while even YOLOv9c gets to 0.7

The author missed 1 setting Suppose I want to train for size 768, 1. In `D-FINE/configs/dfine/include/dfine_hgnetv2.yml`, `eval_spatial_size: [768, 768] # h w` 2. dataloader.yml ```yaml train_dataloader: dataset: transforms: ops: ......

Same here. Only trained after I make classes 0-indexed