aimagelab / meshed-memory-transformer

Meshed-Memory Transformer for Image Captioning. CVPR 2020
BSD 3-Clause "New" or "Revised" License
507 stars 136 forks source link

OSError: Unable to open file (truncated file: eof = 74570219520, sblock->base_addr = 0, stored_eof = 74924166416) #64

Closed HHeracles closed 1 year ago

HHeracles commented 2 years ago

The structure of data directory is: dataset ├── annotations │ ├── captions_train2014.json │ ├── captions_val2014.json │ ├── coco_dev_ids.npy │ ├── IIIT5k_3000 │ ├── coco_restval_ids.npy │ ├── coco_test_ids.npy │ └── coco_train_ids.npy ├── feature │ ├── coco_all_align.hdf5 And to train your model with the parameters used in our experiments, use:

CUDA_VISIBLE_DEVICES=0,1 python train.py --exp_name m2_transformer --batch_size 8 --m 40 --head 8 --warmup 10000 --features_path dataset/feature/coco_all_align.hdf5 --annotation_folder dataset/annotations I get the error as follow: image Why did this error happen? How to solve it?Thank you!

wlcosta commented 1 year ago

This error happened to me because my file was corrupted. From the file's eof and its stored_eof, it appears that your download corrupted the last part of the file. You could try downloading it again.

My MD5 checksum for coco_detections.hdf5 is 695f3eddea88f2f5f7f47febf11e2de2 in case you want to check yours too.