VinVL Image captioning raises AttributeError: 'NoneType' object has no attribute 'img_feature_dim'

[or-toledano]

To reproduce: run the python scripts from VinVL_MODEL_ZOO.md under

Image Captioning on COCO Script to finetune for base model: 1) First train with cross-entropy loss (8 V100 with 16G mem):

And under:

Script to finetune for large model: 1) First train with cross-entropy loss (8 V100 with 32G mem):

Fixed by PR #144

[yaolinli]

To reproduce: run the python scripts from VinVL_MODEL_ZOO.md under

Image Captioning on COCO Script to finetune for base model:

1. First train with cross-entropy loss (8 V100 with 16G mem):

And under:

Script to finetune for large model:

1. First train with cross-entropy loss (8 V100 with 32G mem):

Fixed by PR #144

Hi~ After solving this problem, have you fine-tuned the released ./pretrained_base/checkpoint-2000000 successfully and reproduced the results presented in the paper? Did you successfully download the coco caption dataset used to fine-tune the pre-trained model?

[SYSUmrlian]

my problem is this when runnning oscar Model_zoo.md script it raises:config.img_feature_dim = args.img_feature_dim.AttributeError: 'NoneType' object has no attribute 'img_feature_dim' Please i need help desperately

[or-toledano]

Fixed by PR #144

[SYSUmrlian]

@or-toledano i write down this line: --model_name_or_path pretrained_models/image_captioning/pretrained_base/checkpoint-2000000 but still raise assertionserror: File "oscar/run captioninq.py", line 330, in build dataset assert op.isfile(yaml file)

[densechen]

@SYSUmrlian Try to specify the full path of pretrained model.