Document error reporting. Is there any missing document?

[ccmoe]

Data_io.py file reported an error. Is there any missing file?
IndexError：list index out of range

[nqanh]

There is a special index -1 for the mean frame when the video doesn't have enough frame. This probably causes the error.

[ccmoe]

Can you upload the complete project?Because I really want to do this research. Thank you very much.

[nqanh]

Everything you need to build the code is available :). You can extract your own feature or use my pre-trained resnet50 features. The code should work fine.

[cjiang2]

I think the "data_io.py" file was written earlier before the construction of the main dataset files. (The line in_fid_folder = '/home/anguyen/workspace/dataset/Breakfast/v2c_dataset/subtitle/l30_srt' where I suppose those are just main commands in the "train.txt" and "test.txt" from IIT-V2C dataset)

I've worked on an alternative version of "data_io.py" which will successfully uses the ResNet50 features and constructs the pkl files you need for training and testing. ( "train_resnet50_keras_feature_no_sub_mean.pkl" and "test_resnet50_keras_feature_no_sub_mean.pkl")

# read train/test.txt --> create .pkl file for train and test
import os
import random
import pandas as pd
import numpy as np

random.seed(0)

#in_feature_folder = '/home/anguyen/workspace/dataset/Breakfast/v2c_dataset/video/vgg19_caffe_feature'
#in_feature_folder = '/home/anguyen/workspace/dataset/Breakfast/v2c_dataset/video/resnet50_keras_feature'

feature_name = 'resnet50_keras_feature_no_sub_mean'
#feature_name = 'vgg16_keras_feature_no_sub_mean'
#feature_name = 'inception_keras_feature_no_sub_mean'
#feature_name = 'inception_tensorflow_from_saliency'


#in_feature_folder = '/home/anguyen/workspace/dataset/Breakfast/v2c_dataset/video/resnet50_keras_feature_no_sub_mean'
in_feature_folder = os.path.abspath(os.path.join(os.path.curdir, feature_name))
    
in_train_txt_file = os.path.abspath(os.path.join(os.path.curdir, 'IIT-V2C', 'train.txt'))
in_test_text_file = os.path.abspath(os.path.join(os.path.curdir, 'IIT-V2C', 'test.txt'))

def load_data(in_file):
    data = pd.read_pickle(in_file)
    return data

def load_annotations(dataset_path=os.path.join('IIT-V2C'),
                     annotation_file='train.txt'):
    """Helper function to parse IIT-V2C dataset.
    """
    # Read annotations
    annotations = {}
    with open(os.path.join(dataset_path, annotation_file), 'r') as f:
        i = 0
        annotation = []
        for line in f:
            line = line.strip()
            i += 1
            annotation.append(line)

            if i % 4 == 0:
                # Collect Video Name, Annotated Sub-Video Clip id
                video_fname, command = annotation[0] + '.npy', ' '.join(annotation[2].strip().split(' '))
                annotations[video_fname] = command
                #print(video_fname, command)
                annotation = []
    return annotations

def save_data(list_data, out_file, annotations):
    # faster solution     
    list_video_id = []
    list_caption = []
    for l in list_data:
        print('-------------------------')
        #l = l[0:len(l)-4]  # remove ".npy"
        caption_cmd = annotations[l]
        print('output caption: ', caption_cmd)
        list_caption.append(caption_cmd)
        
        # build full path -- to load later 
        l = in_feature_folder + '/' + l
        list_video_id.append(l)
        print('input video id: ', l)
        
    df = pd.DataFrame({'video_path':list_video_id, 'caption':list_caption})
    print(df)
    
    # save to pickle
    df.to_pickle(out_file)

if __name__ == '__main__':
    
    ### TRAIN TEST FORMAT FILE
    ## video_id_1 caption_1
    ## video_id_2 caption_2
    ## video_id_3 caption_3 

    annotations_train = load_annotations(dataset_path=os.path.join('IIT-V2C'),
                                         annotation_file='train.txt')
    annotations_test = load_annotations(dataset_path=os.path.join('IIT-V2C'),
                                        annotation_file='test.txt')

    list_train = list(annotations_train.keys())
    list_test = list(annotations_test.keys())
    print('LIST TRAIN[0]: ', list_train[0])  #P45_webcam01_P45_juice_22.npy)
    print('LIST TEST[0]: ', list_test[0])
     
    out_folder = os.path.join(os.path.curdir, 'train_test_split')
    if not os.path.exists(out_folder):
        os.makedirs(out_folder)
         
    train_file = out_folder + '/' + 'train_' + feature_name + '.pkl'
    test_file = out_folder + '/' + 'test_' + feature_name + '.pkl'
     
    save_data(list_train, train_file, annotations_train)
    save_data(list_test, test_file, annotations_test)
#     
    
    print('all done!')

[ccmoe]

According to the files provided by the author and zonetrooper 32, it's ready to run. Thank you very much. I'll close this issue.