import torch from model import blocks from model import networks from data_loader.unified_dataloader import UnifiedKeypointDataloader from model_fitting.train import fit import os th_count = 24 dataloader = UnifiedKeypointDataloader(batch_size=6, th_count=th_count) backbone = networks.VGGNetBackbone(inplanes=64, block_counts=[2, 2, 4, 2]) net = networks.OpenPoseNet([backbone], 4, 1, blocks.PoseCNNStage, 10, len(dataloader.trainloader.skeleton) * 2, len(dataloader.trainloader.parts) + 1, dataloader.trainloader.skeleton, dataloader.trainloader.parts) # net = networks.CocoPoseNet() fit(net, dataloader.trainloader, dataloader.validationloader, postprocessing=dataloader.postprocessing, epochs=1000, lower_learning_period=3)
dataset_creator = DatasetCreator(root_dir='./dataset', names_file='data_loader/universalnames.json') for i in range(6): for j in range(5): width = 16 * 2**i depth = 1 + j net = LSTMNet(len(dataset_creator.corpus), width, depth) net.cuda() trainset = dataset_creator.get_train_iterator() trainloader = torch.utils.data.DataLoader(trainset, batch_size=256, shuffle=True, num_workers=th_count, pin_memory=True) validationset = dataset_creator.get_validation_iterator() validationloader = torch.utils.data.DataLoader(validationset, batch_size=1, shuffle=False, num_workers=th_count, pin_memory=False) fit(net, trainloader, validationloader, chp_prefix="{}_{}".format(width, depth), epochs=100, lower_learning_period=10)
trainset = dataset_creator.get_train_iterator() display_iterator_sample(trainset) trainloader = torch.utils.data.DataLoader(trainset, batch_size=16, shuffle=True, num_workers=0) net = ResNet(22) net.cuda() log_datatime = str(datetime.now().time()) loss_writer = SummaryWriter(os.path.join('logs', log_datatime, 'loss')) accuracy_writer = SummaryWriter(os.path.join('logs', log_datatime, 'accuracy')) validationset = dataset_creator.get_validation_iterator() validationloader = torch.utils.data.DataLoader(validationset, batch_size=32, shuffle=False, num_workers=0) fit(net, trainloader, validationloader, loss_writer, accuracy_writer, trainset.labels, epochs=100) loss_writer.close() accuracy_writer.close()
import torch from model import blocks from model import networks from data_loader.unified_dataloader import UnifiedDataloader from model_fitting.train import fit import os th_count = 24 dataloader = UnifiedDataloader(batch_size=32, th_count=th_count) net = networks.AoANet(512, dataloader.vectorizer) net.grad_backbone(False) fit(net, dataloader.trainloader, dataloader.validationloader, epochs=1000, lower_learning_period=2)
import torch from model import blocks from model import networks from data_loader.dataset_provider import SegmentationDatasetProvider from model_fitting.train import fit, test import os th_count = 12 dataset_name = 'custom_car' net_backbone = networks.ResNetBackbone(block=blocks.BasicBlock, block_counts=[3, 4, 6], inplanes=64) net = networks.DeepLabV3Plus(net_backbone, 2) data_provider = SegmentationDatasetProvider(net, batch_size=4, th_count=th_count) fit(net, data_provider.trainloader, data_provider.validationloader, dataset_name=dataset_name, epochs=1000, lower_learning_period=5) test(net, data_provider.testloader, dataset_name=dataset_name)
(51, 'bowl'), (52, 'banana'), (53, 'apple'), (54, 'sandwich'), (55, 'orange'), (56, 'broccoli'), (57, 'carrot'), (58, 'hot dog'), (59, 'pizza'), (60, 'donut'), (61, 'cake'), (62, 'chair'), (63, 'couch'), (64, 'potted plant'), (65, 'bed'), (67, 'dining table'), (70, 'toilet'), (72, 'tv'), (73, 'laptop'), (74, 'mouse'), (75, 'remote'), (76, 'keyboard'), (77, 'cell phone'), (78, 'microwave'), (79, 'oven'), (80, 'toaster'), (81, 'sink'), (82, 'refrigerator'), (84, 'book'), (85, 'clock'), (86, 'vase'), (87, 'scissors'), (88, 'teddy bear'), (89, 'hair drier'), (90, 'toothbrush')] backbone = networks.ResNetBackbone(inplanes=64, block=blocks.BasicBlock, block_counts=[3, 4, 6, 3]) net = networks.RetinaNet(backbone=[networks.FeaturePyramidBackbone, backbone], classes=classes, ratios=ratios) coco_provider = CocoDetectionDatasetProvider(net, annDir=os.path.join( '/Data', dataset_name), batch_size=8, th_count=th_count) fit(net, coco_provider.trainloader, coco_provider.validationloader, dataset_name=dataset_name, box_transform=coco_provider.target_to_box_transform, epochs=1000, lower_learning_period=3)