Closed dmandreev closed 6 years ago
It supports. you can use video-classification-3d-cnn-pytorch to extract c3d features, then mean pool to get 2048 dim feature for each video
Hi! I do used the repo video-classification-3d-cnn-pytorch, but its output is a json. How can I use that with your codes?
Thank you for your work!
I used this to turn the json into npy:
import json
import numpy as np
import os
videos = json.load(open('./output.json', 'r'))
if not os.path.exists("data/3dfeature/"):
os.makedirs("data/3dfeature/")
for video in videos:
name = str(video["video"].split(".")[0])
segments = video["clips"]
feat = []
for segment in segments:
feat += segment["features"]
np.save("data/3dfeature/" + name + ".npy", feat)
Hello! Thank you for your work! Is c3d features currently not supported? I saw some traces in sources, but not sure. Could you please give any hints? Is these features have to be extracted as resnet features and later simply jointed?