# compute validation meanIoU
val_IoUs.append(compute_avg_IoU(**val_data))
# compute test meanIoU
test_IoUs.append(compute_avg_IoU(**test_data))
with tables.openFile('ious_kth{0:02d}left_out.h5'.format(args.test_person),
'w') as h5file:
h5file.createArray(h5file.root, 'train_IoUs', np.array(train_IoUs))
h5file.createArray(h5file.root, 'val_IoUs', np.array(val_IoUs))
h5file.createArray(h5file.root, 'test_IoUs', np.array(test_IoUs))
h5file.createArray(h5file.root, 'trainer_costs', trainer.costs)
if best_val_IoU < val_IoUs[-1]:
best_val_IoU = val_IoUs[-1]
best_val_IoU_epoch = epoch + 1
model.save('attention_model_kth_{0:02d}left_out_val_best.h5'.format(
args.test_person))
print 'train IoU: {0:f}'.format(train_IoUs[-1])
print 'val IoU: {0:f} (best so far {1} in epoch {2})'.format(
val_IoUs[-1], best_val_IoU, best_val_IoU_epoch)
print 'test IoU: {0:f}'.format(test_IoUs[-1])
plt.clf()
plt.plot(range(epoch + 2), train_IoUs, c='g', label='Avg. IoU on Train')
plt.plot(range(epoch + 2),
val_IoUs,
c='b',
label='Avg. IoU on Val (person {0})'.format(val_person))
plt.scatter([best_val_IoU_epoch], [best_val_IoU], marker='*', c='b')
plt.xlabel('epoch')
plt.ylabel('avg. IoU')
val_IoUs.append(compute_avg_IoU(**val_data))
# compute test meanIoU
test_IoUs.append(compute_avg_IoU(**test_data))
with tables.openFile('ious_kth{0:02d}left_out.h5'.format(args.test_person),
'w') as h5file:
h5file.createArray(h5file.root, 'train_IoUs', np.array(train_IoUs))
h5file.createArray(h5file.root, 'val_IoUs', np.array(val_IoUs))
h5file.createArray(h5file.root, 'test_IoUs', np.array(test_IoUs))
h5file.createArray(h5file.root, 'trainer_costs', trainer.costs)
if best_val_IoU < val_IoUs[-1]:
best_val_IoU = val_IoUs[-1]
best_val_IoU_epoch = epoch + 1
model.save(
'attention_model_kth_{0:02d}left_out_val_best.h5'.format(
args.test_person))
print 'train IoU: {0:f}'.format(train_IoUs[-1])
print 'val IoU: {0:f} (best so far {1} in epoch {2})'.format(
val_IoUs[-1], best_val_IoU, best_val_IoU_epoch)
print 'test IoU: {0:f}'.format(test_IoUs[-1])
plt.clf()
plt.plot(range(epoch + 2), train_IoUs, c='g',
label='Avg. IoU on Train')
plt.plot(range(epoch + 2), val_IoUs, c='b',
label='Avg. IoU on Val (person {0})'.format(val_person))
plt.scatter([best_val_IoU_epoch], [best_val_IoU],
marker='*', c='b')
plt.xlabel('epoch')
