class TestDataset(Dataset): def __init__(self, config, split='test'): self.config = config if config.test_data_dir: self.db = WaterDataset(config.test_data_dir, split=split) else: self.db = WaterDataset(config.data_dir, split=split) #self.db = aaaDataset(config.data_dir, config.num_of_samples, split=split) def __getitem__(self, idx): label, datas = self.db.get_example(idx) #label, datas = self.db.get_example(idx, 'test') #label = t.from_numpy(np.array(label).astype(np.float32)) label = t.from_numpy(np.array(label)) datas = np.array(datas) datas = t.from_numpy(datas) datas = datas.contiguous().view(1, -1) # TODO: check whose stride is negative to fix this instead copy all return label, datas def __len__(self): return len(self.db)
class TrainDataset(Dataset): def __init__(self, config, split='train'): self.config = config self.db = WaterDataset(config.data_dir, split=split) def __getitem__(self, idx): label, datas = self.db.get_example(idx) label = t.from_numpy(np.array(label)) datas = np.array(datas) datas = t.from_numpy(datas) datas = datas.contiguous().view(1, -1) # TODO: check whose stride is negative to fix this instead copy all return label, datas def __len__(self): return len(self.db)
def __init__(self, config, split='test'): self.config = config self.db = WaterDataset(config.data_dir, split=split)
def __init__(self, config, split='test'): self.config = config if config.test_data_dir: self.db = WaterDataset(config.test_data_dir, split=split) else: self.db = WaterDataset(config.data_dir, split=split)
def __init__(self, config, split='train'): self.config = config print(config.data_dir) self.db = WaterDataset(config.data_dir, config.num_of_samples, split=split)
# Normalize by softmax all_E[:, :, f + 1] = torch.clamp(all_E[:, :, f + 1], 1e-7, 1 - 1e-7) all_E[:, :, f + 1] = torch.log( (all_E[:, :, f + 1] / (1 - all_E[:, :, f + 1]))) all_E[:, :, f + 1] = F.softmax(Variable(all_E[:, :, f + 1]), dim=1).data return all_E if MO: Testset = DAVIS(DAVIS_ROOT, imset='2017/val.txt', multi_object=True) if SO: Testset = DAVIS(DAVIS_ROOT, imset='2016/val.txt') if test_case is not None: Testset = WaterDataset(water_root, test_case) Testloader = data.DataLoader(Testset, batch_size=1, shuffle=False, num_workers=2, pin_memory=True) model = nn.DataParallel(RGMP()) if torch.cuda.is_available(): model.cuda() model.load_state_dict(torch.load('weights.pth')) model.eval() # turn-off BN for seq, (all_F, all_M, info) in enumerate(Testloader): all_F, all_M = all_F[0], all_M[0]