WIP mb/image2pulsetrain #54
Conversation
| const_amp=20, const_freq=20, rftype='gaussian', rfsize=None, | ||
| tsample=0.005 / 1000, dur=0.5, pulsedur=0.5 / 1000., | ||
| interphasedur=0.5 / 1000., pulsetype='cathodicfirst'): | ||
| try: |
There was a problem hiding this comment.
Maybe only try to import skimage if a string is provided, or if the image is RGB? We could allow a simple gray-scale input, if people want to provide that directly.
There was a problem hiding this comment.
Agree, except we always need it for resizing the image accordingly... (see below)
| # Convert image to grayscale | ||
| img_large = rgb2gray(np.array(img)).astype(np.float32) | ||
|
|
||
| # Maximize contrast |
There was a problem hiding this comment.
Wouldn't this lead to confusing results in an experiment in which different contrasts are used?
There was a problem hiding this comment.
Good point. Added an input argument max_contrast so users can choose.
| yhi = np.max(xyr[:, 1] + xyr[:, 2]) | ||
|
|
||
| # Resize the image accordingly and flip up down | ||
| img_resize = resize(img_large, (yhi - ylo, xhi - xlo)) |
There was a problem hiding this comment.
I guess we'll always need skimage for this...
| for e in implant: | ||
| rf = receptive_field(e, xg, yg, rftype, rfsize) | ||
| magn.append(np.mean(rf * img_resize)) | ||
| magn = np.array(magn) |
There was a problem hiding this comment.
We might be able to preallocate this one.
| e_s += "'frequency'." | ||
| raise ValueError(e_s) | ||
|
|
||
| pt = Psycho2Pulsetrain(tsample, freq=freq, amp=amp, dur=dur, |
There was a problem hiding this comment.
Maybe there's a way to speed up/save memory by doing these all in the same time-series object? OTOH, might not be worth the hassle.
Work in progress: Add functionality to encode an image as a list of pulse trains.