In [ ]:

import os
import numpy as np
import matplotlib.pyplot as plt

In [ ]:

# @title Download the data

import os, requests, tarfile

fnames = ["kay_labels.npy", "kay_labels_val.npy", "kay_images.npz"]
urls = ["https://osf.io/r638s/download",
        "https://osf.io/yqb3e/download",
        "https://osf.io/ymnjv/download"]

for fname, url in zip(fnames, urls):
  if not os.path.isfile(fname):
    try:
      r = requests.get(url)
    except requests.ConnectionError:
      print("!!! Failed to download data !!!")
    else:
      if r.status_code != requests.codes.ok:
        print("!!! Failed to download data !!!")
      else:
        print(f"Downloading {fname}...")
        with open(fname, "wb") as fid:
          fid.write(r.content)
        print(f"Download {fname} completed!")

Downloading kay_labels.npy...
Download kay_labels.npy completed!
Downloading kay_labels_val.npy...
Download kay_labels_val.npy completed!
Downloading kay_images.npz...
Download kay_images.npz completed!

In [ ]:

with np.load(fname) as dobj:
  dat = dict(**dobj)
labels = np.load('kay_labels.npy')
val_labels = np.load('kay_labels_val.npy')

labels is a 4 by stim array of class names:

row 3 has the labels predicted by a deep neural network (DNN) trained on Imagenet
rows 0-2 correspond to different levels of the wordnet hierarchy for the DNN predictions

In [ ]:

print(labels[:, :4])

[['artifact' 'entity' 'animal' 'animal']
 ['instrumentality' 'round shape' 'vertebrate' 'vertebrate']
 ['equipment' 'sphere' 'mammal' 'amphibian']
 ['croquet ball' 'bubble' 'komondor' 'tailed frog']]

In [ ]:

print(dat.keys())

dict_keys(['stimuli', 'stimuli_test', 'responses', 'responses_test', 'roi', 'roi_names'])

dat has the following fields:

stimuli: stim x i x j array of grayscale stimulus images
stimuli_test: stim x i x j array of grayscale stimulus images in the test set
responses: stim x voxel array of z-scored BOLD response amplitude
responses_test: stim x voxel array of z-scored BOLD response amplitude in the test set
roi: array of voxel labels
roi_names: array of names corresponding to voxel labels

In [ ]:

print(dat["stimuli"].shape)

(1750, 128, 128)

In [ ]:

print(dat["responses"].shape)

(1750, 8428)

This is the number of voxels in each ROI. Note that "Other" voxels have been removed from this version of the dataset:

In [ ]:

dict(zip(dat["roi_names"], np.bincount(dat["roi"])))

Out[ ]:

{'LatOcc': 928,
 'Other': 0,
 'V1': 1294,
 'V2': 2083,
 'V3': 1790,
 'V3A': 484,
 'V3B': 314,
 'V4': 1535}

Each stimulus is a 128 x 128 grayscale array:

In [ ]:

fig, axs = plt.subplots(2, 4, figsize=(12, 6), sharex=True, sharey=True)
for ax, im, lbl in zip(axs.flat, dat["stimuli"], labels[-1, :]):
  ax.imshow(im, cmap="gray")
  ax.set_title(lbl)
fig.tight_layout()
fig.show()

Each stimulus is associated with a pattern of BOLD response across voxels in visual cortex:

In [ ]:

fig, ax = plt.subplots(figsize=(12, 5))
ax.set(xlabel="Voxel", ylabel="Stimulus")
heatmap = ax.imshow(dat["responses"],
                    aspect="auto", vmin=-1, vmax=1, cmap="bwr")
fig.colorbar(heatmap, shrink=.5, label="Response amplitude (Z)")
fig.tight_layout()
fig.show()