gulvarol/bsl1k

Missing Poses and Duplicated for MSASL

Opened this issue · 0 comments

Hello,

Thank you so much for the repo, it is really helpful.

I am trying to use the skeleton(pose) keypoints provided for MSASL.
The MSASL should contain 25513 videos.
I found the issues mentioned below:

  1. there are missing skeleton (pose : None ). Hence, the skeleton information is only available for 25147 videos.
  2. From those 25417 skeletons there are duplicates with respect to the video ID.

This means that the skeleton information is actually available for 25059.
Could you please help me with that ?

The code that I used is :

`

import numpy as np
import time 
import pickle
import json 
import sys
import os
path = './info/pose.pkl'
lpath = './info/info.pkl'
label_data = pickle.load(open(lpath, "rb"))
print(label_data.keys())

name_data=label_data['videos']["name"]
video_data=label_data['videos']["word"]
labelID_data=label_data['videos']["word_id"]
split=label_data['videos']['split']

pose_data = pickle.load(open(path, "rb"))

print("missing")
k = 0 
for i in  pose_data["pose"]:
    if i is None : 
        k = k+1
print('There are ',k, 'None poses ')


print(create dictionary in order to find if there are duplicated )
data = {}
print(len(pose_data["pose"])) # 25439

for i in range(len(pose_data["pose"])):
    data_numpy = pose_data["pose"][i]
    # print('*************')
    if data_numpy is None : 
        continue
    if name_data[i] in data:
        print("already exists ", name_data[i] )
        print(data[name_data[i]] ,  split[i] )
        # time.sleep(5)

    data[name_data[i]] = split[i]

print('Number of videos : ', len(data))

`