search

ExID-proj / VILBERT_tutorial

A step-by-step tutorial on how to use VILBERT to fine-tune and test on new data, and how to identify the visiolinguistic embeddings.
6 stars 0 forks source link

Applying on other datasets than flickr30 dataset error #2

Closed Souryadipstan closed 2 years ago

Souryadipstan commented 2 years ago

Traceback (most recent call last): File "/Users/sourya/Desktop/vilbert-multi-task/identify_vilbert_emds.py", line 220, in main() File "/Users/sourya/Desktop/vilbert-multi-task/identify_vilbert_emds.py", line 147, in main task_batch_size, task_num_iters, task_id, task_datasets_val, task_dataloader_val = LoadDatasetTransfer( File "/Users/sourya/Desktop/vilbert-multi-task/vilbert/task_utils.py", line 408, in LoadDatasetTransfer task_datasets_val[task] = DatasetMapTrans[task_name]( File "/Users/sourya/Desktop/vilbert-multi-task/vilbert/datasets/retreival_dataset.py", line 494, in init features, numboxes, boxes, = self._image_features_reader[image_id] File "/Users/sourya/Desktop/vilbert-multi-task/vilbert/datasets/_image_features_reader.py", line 71, in getitem index = self._image_ids.index(image_id) ValueError: b'61' is not in list

This error is showing up when I am trying to generate the Visiolinguistic Embedding. I am using my own dataset. Previously, another bug showed up because, my file names were not integer but strings unlike the flickr30 dataset so, I converted the filenames to integer. After that, this bug is occurring. Any workaround?

Souryadipstan commented 2 years ago

Found out the error occurred as I renamed the file after creating the features from the images. File names are also saved as image_id's in the .npy file so, renaming the files first then again generating features fixed the bug.