Video Classification

yanagar25 · February 18, 2018, 1:52pm

Hi,

I’m trying to predict 4 classes of video dataset. my videos are composed from 32 frames saved as images on my hard disk.
The dataset is structured as follows:
Train Folder:
- Label 1:
- Video 1 Folder:
- img1.jpg
- img2.jpg

I tried writing my own VideoClassifierData which is implementing all the methods in BaseDataset. (if needed i will post my entire code of the VideoClassifierData)
My dataset works, it returns a list of numpy images of the frames as a video.
When i’m passing my VideoClassifierData to ConvLearner and fitting the model, i’m getting an error that the output is not 1 dimensional array when trying to create a variable in model.py (loss = stepper.step(V(x),V(y)))
Anyone has insight regarding this issue? or maybe direct me to what i’m doing wrong?

Yana

QuantScientist · February 18, 2018, 2:23pm

Do upload your full code to git with a sample DB.
If you are looking to load real video, then you can use this:

github.com

MohsenFayyaz89/PyTorch_Video_Dataset/blob/master/videoDataset.py

"""
PyTorch Video Dataset Class for loading videos using PyTorch
Dataloader. This Dataset assumes that video files are Preprocessed
 by being trimmed over time and resizing the frames.

Mohsen Fayyaz __ Sensifai Vision Group
http://www.Sensifai.com

If you find this code useful, please star the repository.
"""

from __future__ import print_function, division
import cv2
import os
import torch
import numpy as np
import pickle
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms, utils

This file has been truncated. show original

yanagar25 · February 19, 2018, 7:14am

Hi,
Thanks for the reply.
First, i’m using the fastai framework to train my network so I built my own VideoClassifierData similar to ImageClassifierData that just loading video instead of single image, that’s why the code above not really helps me in this current situation.
Second, ere you can see the problem i’m having: https://github.com/yana25/video_classification
just install the fast.ai library using the link in the readme and run “fastai_video_train.py” and see the error i’m
getting.

Yana