This code downsamples the video to a width of resizewidth. The video is extracted as a numpy array, which is then clustered with kmeans, whereby each frames is treated as a vector. Frames from different clusters are then selected for labeling. This procedure makes sure that the frames "
(
clip,
numframes2pick,
start,
stop,
Index=None,
step=1,
resizewidth=30,
batchsize=100,
max_iter=50,
color=False,
)
| 108 | |
| 109 | |
| 110 | def KmeansbasedFrameselection( |
| 111 | clip, |
| 112 | numframes2pick, |
| 113 | start, |
| 114 | stop, |
| 115 | Index=None, |
| 116 | step=1, |
| 117 | resizewidth=30, |
| 118 | batchsize=100, |
| 119 | max_iter=50, |
| 120 | color=False, |
| 121 | ): |
| 122 | """This code downsamples the video to a width of resizewidth. |
| 123 | |
| 124 | The video is extracted as a numpy array, which is then clustered with kmeans, whereby each frames is treated as a |
| 125 | vector. |
| 126 | Frames from different clusters are then selected for labeling. This procedure makes sure that the frames "look |
| 127 | different", |
| 128 | i.e. different postures etc. On large videos this code is slow. |
| 129 | |
| 130 | Consider not extracting the frames from the whole video but rather set start and stop to a period around interesting |
| 131 | behavior. |
| 132 | |
| 133 | Note: this method can return fewer images than numframes2pick. |
| 134 | """ |
| 135 | |
| 136 | print( |
| 137 | "Kmeans-quantization based extracting of frames from", |
| 138 | round(start * clip.duration, 2), |
| 139 | " seconds to", |
| 140 | round(stop * clip.duration, 2), |
| 141 | " seconds.", |
| 142 | ) |
| 143 | startindex = int(np.floor(clip.fps * clip.duration * start)) |
| 144 | stopindex = int(np.ceil(clip.fps * clip.duration * stop)) |
| 145 | |
| 146 | if Index is None: |
| 147 | Index = np.arange(startindex, stopindex, step) |
| 148 | else: |
| 149 | Index = np.array(Index) |
| 150 | Index = Index[(Index > startindex) * (Index < stopindex)] # crop to range! |
| 151 | |
| 152 | nframes = len(Index) |
| 153 | if batchsize > nframes: |
| 154 | batchsize = int(nframes / 2) |
| 155 | |
| 156 | if len(Index) >= numframes2pick: |
| 157 | clipresized = clip.resize(width=resizewidth) |
| 158 | ny, nx = clipresized.size |
| 159 | frame0 = img_as_ubyte(clip.get_frame(0)) |
| 160 | if np.ndim(frame0) == 3: |
| 161 | ncolors = np.shape(frame0)[2] |
| 162 | else: |
| 163 | ncolors = 1 |
| 164 | print("Extracting and downsampling...", nframes, " frames from the video.") |
| 165 | |
| 166 | if color and ncolors > 1: |
| 167 | DATA = np.zeros((nframes, nx * 3, ny)) |