train=False)
idx = opt.idx
print("model %d/%d" % (idx, len(d)))
point, seg = d[idx]
print(point.shape, seg.shape)
point_np = point.asnumpy()
ctx = mx.gpu()
cmap = plt.cm.get_cmap("hsv", 10)
cmap = np.array([cmap(i) for i in range(10)])[:, :3]
gt = cmap[seg.asnumpy().astype(np.uint8) - 1, :]
classifier = PointNetDenseCls(k=4)
classifier.load_parameters(opt.model, ctx=ctx)
point = nd.expand_dims(point.transpose((1, 0)), axis=0)
pred, _ = classifier(point.as_in_context(ctx))
pred_choice = pred.argmax(2)
print(pred_choice)
#print(pred_choice.size())
pred_color = cmap[pred_choice.asnumpy().astype(np.uint8)[0], :]
#print(pred_color.shape)
showpoints(point_np, gt, pred_color)
print(len(dataset), len(test_dataset))
num_classes = dataset.num_seg_classes
print('classes', num_classes)
try:
os.makedirs(opt.outf)
except OSError:
pass
blue = lambda x: '\033[94m' + x + '\033[0m'
ctx = mx.gpu()
classifier = PointNetDenseCls(k=num_classes, routing=1)
classifier.initialize(ctx=ctx)
if opt.model != '':
classifier.load_parameters(opt.model)
optimizer = Trainer(classifier.collect_params(), 'sgd', {
'learning_rate': 0.001,
'momentum': 0.9
})
L_loss = loss.SoftmaxCrossEntropyLoss()
num_batch = len(dataset) / opt.batchSize
for epoch in range(opt.nepoch):
correct = 0.
L = 0.
count = 0
L_eval = 0.
correct_eval = 0.