FashionBERT pretrained model error

Hello,

Thanks for sharing the code and the pretrained model. But I get the invalid argument error as follows while using it.
)
[link_to_error_image](https://user-images.githubusercontent.com/70706034/127808689-0d5220b5-1eb0-4b78-ac8c-e62b9c6c97ec.jpg)

I am using the following script:
`python pretrain_main.py \
--workerGPU=1 \
--type=img2txt \
--mode=predict \
--predict_input_fp=eval_img2txt_list.list_csv \
--predict_batch_size=64 \
--input_sequence_length=64 \
--output_dir=./fashionbert_out \
--pretrain_model_name_or_path=fashionbert_pretrain_model_fin \
--image_feature_size=131072 \
--predict_checkpoint_path=./fashionbert_pretrain_model_fin/model.ckpt-54198 \
--input_schema="image_feature:float:131072,image_mask:int:64,input_ids:int:64,input_mask:int:64,segment_ids:int:64,nx_sent_labels:int:1,prod_desc:str:1,text_prod_id:str:1,image_prod_id:str:1,prod_img_id:str:1" \`

Thanks & Regards
@dannygao1984 @minghui

Start with pretrain_main.py and reproduce the provided prediction command using fashionbert_pretrain_model_fin and model.ckpt-54198. Inspect the linked error image and the command's input_schema and image_feature settings; done means the FashionBERT img2txt prediction runs without the invalid-argument error.

Written by the indexing model from the issue text.