Pytables demo to store tracks of a video database

tracks_pytables_demo.py

#!/usr/bin/env python

import numpy as np
import tables

# class containing infos to identify a track
class Track(tables.IsDescription):
    id = tables.IntCol()      # track ID (should be unique)
    frame = tables.IntCol()   # frame where the current track bounding box was found
    ulx = tables.IntCol()     # upper left corner's x coordinate of the bbox (upper left is origin (0,0) of frame)
    uly = tables.IntCol()     # upper left corner's y coordinate of the bbox
    width = tables.IntCol()   # width of the bbox
    height = tables.IntCol()  # height of the bbox
    score = tables.FloatCol() # detector score of the bbox (if not found manually)
    
# create dummy data
np.random.seed(1)
movies = {}
movies['train'] = [ 'tr_movie%d' % i for i in range(1000) ]
movies['test']  = [ 'te_movie%d' % i for i in range(1000) ]
sids = {}
for s in "train test".split():
    sids[s] = {}
    for m in movies[s]:
        sids[s][m] = ( '%s-frames-%d-%d' % \
            (m, np.random.randint(100), np.random.randint(100, 200))
            for i in range(10) )

print "beginning writing .h5 file"

filename = 'test.h5'
# Open a file in "w"rite mode
h5file = tables.openFile(filename, mode = "w", title = "hollywood2 dataset")
# Create the tracks group at the root
tracks_g = h5file.createGroup(h5file.root, 'tracks', 'Tracks for the dataset')
# Create the training and test groups
trt_g = h5file.createGroup(tracks_g, 'train', 'Tracks for the training samples')
tet_g = h5file.createGroup(tracks_g, 'test', 'Tracks for the test samples')
# Create the automatic and ground truth groups of tracks in each case
gnames  = ['autom', 'gt']
gtitles = ['Automatically detected tracks', 'Ground truth tracks']
for s in "train test".split():
    for gname, gtitle in zip(gnames, gtitles):
        ttr = h5file.createGroup("/tracks/%s" % s, gname, gtitle)
# Create the per-movie tables of tracks
tid = -1
for s in "train test".split():
    for t in "autom gt".split():
        for m in movies[s]:
            pg = "/tracks/%s/%s" % (s,t)
            table = h5file.createTable(pg, m, Track, "Movie %s" % m)
            trow = table.row
            # add the sample tracks
            for sid in sids[s][m]:
                # add multiple tracks
                for atn in range(10):
                    tid += 1
                    trow['id'] = tid
                    trow['frame'] = np.random.randint(100)
                    trow['ulx'] = np.random.randint(200)
                    trow['uly'] = np.random.randint(100)
                    trow['width'] = np.random.randint(120)
                    trow['height'] = np.random.randint(140)
                    trow['score'] = np.random.random()
                    trow.append()
            table.flush()

# Close (and flush) the file
h5file.close()

print ".h5 written"

# open the file and display its contents
ff = tables.openFile("test.h5")

#ff # prints all infos in interactive session
print ff.root
print ff.root.tracks
print ff.root.tracks.train
print ff.root.tracks.train.autom
print ff.root.tracks.train.getattr('autom')
mv1 = ff.root.tracks.train.autom.tr_movie1
print mv1
print mv1[:3]
print mv1.cols.frame[:3]
print [ row['frame'] for row in mv1.where('(score >= 0.5) & (width > 25)') ]

ff.close()
raw_input()