Open 666wodeyy opened 5 months ago
Also met this error.
same issue here
I finally solved this problem by modifying my configuration file and redownload the official project, since I'd done some modifications to the source codes.
获取 Outlook for iOShttps://aka.ms/o0ukef
发件人: Donghaowang @.> 发送时间: Tuesday, August 27, 2024 10:52:51 AM 收件人: open-mmlab/mmdetection @.> 抄送: Cloud65000 @.>; Comment @.> 主题: Re: [open-mmlab/mmdetection] TypeError: unhashable type: 'dict' (Issue #11645)
same issue here
― Reply to this email directly, view it on GitHubhttps://github.com/open-mmlab/mmdetection/issues/11645#issuecomment-2311467629, or unsubscribehttps://github.com/notifications/unsubscribe-auth/BDEN2RP5N62374TH53JTJP3ZTPSYHAVCNFSM6AAAAABGMQSFY2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGMJRGQ3DONRSHE. You are receiving this because you commented.Message ID: @.***>
when i try to finrtune mm-grounding dion with my custom dataset this error occur:
the "text_prompts" is :
it's form :
text_prompts = [ data_samples.text for data_samples in batch_data_samples ]
one batch_data_samples.text is :
my data is defined in cfg like this :
train_dataloader = dict( batch_size=3, num_workers=4, dataset=dict( _delete_=True, type='ODVGDataset', backend_args=None, data_root=data_root, metainfo=metainfo, return_classes=True, pipeline=train_pipeline, label_map_file='path/to/my label map', filter_cfg=dict(filter_empty_gt=False), ann_file='path/to/my/anno ', data_prefix=dict(img=' ')))