About Train Part 1

[HuangZhe885]

Thank you very much for sharing your great work. I've encountered some challenges while trying to train using different datasets.
When I use dateset 'dair_v2x' to run,
python opencood/tools/train.py --hypes_yaml opencood/hypes_yaml/dair-v2x/npj/dair_where2comm_max_multiscale_resnet.yaml
I noticed that the 'past_k_time_interval' keyword is not present in the 'IntermediateFusionDatasetDAIRIrregular' function.

And when I use dateset 'opv2v,'
python opencood/tools/train.py --hypes_yaml /opencood/hypes_yaml/opv2v/npj/opv2v_irr_past_where2comm_max_multiscale_resnet.yaml
I found that there is no 'timestamps.npy' in the raw data. Do I need to regenerate these timestamps, How to regenerate these?
Once again, thank you for your excellent work!

[SizheWei]

@HuangZhe885 Thank you for your interest and support!

In response to your inquiries:

For the first issue, please review the yaml file again to ensure the presence of the 'past_k_time_interval' keyword for the DAIR-V2X dataset. If the problem persists, kindly provide details about the dataset and the specific SOTA method you are using.

Regarding the second issue, the 'opv2v_irr' yaml is used for our new dataset IRV2V, not the OPV2V dataset. We are currently preparing tutorials for this new dataset to facilitate researchers' use. Once the tutorials are finalized and the IRV2V dataset is released, we will update this repository accordingly. Thank you for your patience and understanding.

Much appreciate for your support and encouragement!

[HuangZhe885]

I use this yaml file to generate [Encoder, Decoder, and ROI Generator]
opencood/hypes_yaml/dair-v2x/npj/dair_where2comm_max_multiscale_resnet.yaml
and it called this fusion method "IntermediateFusionDatasetDAIRIrregular",
'past_k_time_interval' keyword is not exits in dict:

    selected_cav_processed.update({'object_bbx_center': object_bbx_center[object_bbx_mask == 1],
         'object_ids': object_ids,
         'projected_lidar': projected_lidar,
         'processed_features': processed_lidar,
         'transformation_matrix': transformation_matrix,
         'transformation_matrix_clean': transformation_matrix_clean})

Structure of dataset is : and replace by new annotations files and data.

         |-- cooperative-vehicle-infrastructure 
            |-- vehicle-side
            |-- infrastructure-side
            |-- cooperative
            |-- val.json
            |-- train.json

thank you for your kindly response.

[HuangZhe885]

I use this yaml file to generate [Encoder, Decoder, and ROI Generator] opencood/hypes_yaml/dair-v2x/npj/dair_where2comm_max_multiscale_resnet.yaml and it called this fusion method "IntermediateFusionDatasetDAIRIrregular", 'past_k_time_interval' keyword is not exits in dict:

    selected_cav_processed.update({'object_bbx_center': object_bbx_center[object_bbx_mask == 1],
         'object_ids': object_ids,
         'projected_lidar': projected_lidar,
         'processed_features': processed_lidar,
         'transformation_matrix': transformation_matrix,
         'transformation_matrix_clean': transformation_matrix_clean})

Structure of dataset is : and replace by new annotations files and data.

         |-- cooperative-vehicle-infrastructure 
            |-- vehicle-side
            |-- infrastructure-side
            |-- cooperative
            |-- val.json
            |-- train.json

thank you for your kindly response.

By the way, If an optimal weight net_epoch_bestval_at22.pth is generated in the first stage, how to use this weight in fornewdataset.py? I don't understand what scenario_database and debug_path mean?

[scz023]

我使用这个 yaml 文件生成 [编码器、解码器和 ROI 生成器]，它将这种融合方法称为“IntermediateFusionDatasetDAIRIrregular”，“past_k_time_interval”关键字不在字典中退出：opencood/hypes_yaml/dair-v2x/npj/dair_where2comm_max_multiscale_resnet.yaml

    selected_cav_processed.update({'object_bbx_center': object_bbx_center[object_bbx_mask == 1],
         'object_ids': object_ids,
         'projected_lidar': projected_lidar,
         'processed_features': processed_lidar,
         'transformation_matrix': transformation_matrix,
         'transformation_matrix_clean': transformation_matrix_clean})

数据集的结构是：，并用新的注释文件和数据替换。

         |-- cooperative-vehicle-infrastructure 
            |-- vehicle-side
            |-- infrastructure-side
            |-- cooperative
            |-- val.json
            |-- train.json

感谢您的友好回复。

I meet the same problem
FileNotFoundError: [Errno 2] No such file or directory: '/home/scz/hetecooper/dataset/DAIR-V2X/DAIR_V2X_C/.fr-8b3J8m/timestamps.npy'
even after replaced the annotation fields, did you solved it？and how did you solved it?