def pixelwise():
msrc = MSRCDataset()
train = msrc.get_split('train')
predictions = []
for filename in train:
probs = load_kraehenbuehl(filename)
prediction = np.argmax(probs, axis=-1)
predictions.append(prediction)
msrc.eval_pixel_performance(train, predictions)
def pixelwise():
msrc = MSRCDataset()
train = msrc.get_split('train')
predictions = []
for filename in train:
probs = load_kraehenbuehl(filename)
prediction = np.argmax(probs, axis=-1)
predictions.append(prediction)
msrc.eval_pixel_performance(train, predictions)
def main():
msrc = MSRCDataset()
images = msrc.get_split()
for image_name in images:
image = msrc.get_image(image_name)
fig, axes = plt.subplots(1, 4, figsize=(12, 6))
axes[0].imshow(image)
axes[1].set_title("ground truth")
axes[1].imshow(image)
gt = msrc.get_ground_truth(image_name)
axes[1].imshow(colors[gt], alpha=.7)
axes[2].set_title("new ground_truth")
gt_new = msrc.get_ground_truth(image_name, ds="new")
axes[2].imshow(image)
axes[2].imshow(colors[gt_new], vmin=0, vmax=23, alpha=.7)
present_y = np.unique(np.hstack([gt.ravel(), gt_new.ravel()]))
axes[3].imshow(colors[present_y, :][:, np.newaxis, :],
interpolation='nearest')
for i, c in enumerate(present_y):
axes[3].text(1, i, classes[c])
for ax in axes.ravel():
ax.set_xticks(())
ax.set_yticks(())
fig.savefig("new_gt/%s.png" % image_name, bbox_inches="tight")
plt.close(fig)