data: root: '/root/autodl-fs/DATASET/MapLocNetDataset/UAV/' train_citys: - Paris - Berlin - London - Tokyo - NewYork val_citys: # - Taipei # - LosAngeles # - Singapore - SanFrancisco test_citys: - SanFrancisco image_size: 256 train: batch_size: 12 num_workers: 4 val: batch_size: ${..train.batch_size} num_workers: ${.batch_size} num_classes: areas: 7 ways: 10 nodes: 33 pixel_per_meter: 1 crop_size_meters: 64 max_init_error: 48 add_map_mask: true resize_image: 512 pad_to_square: true rectify_pitch: true augmentation: rot90: true flip: true image: apply: True brightness: 0.5 contrast: 0.4 saturation: 0.4 hue": 0.5/3.14 model: image_size: ${data.image_size} latent_dim: 128 val_citys: ${data.val_citys} image_encoder: name: feature_extractor_v2 backbone: encoder: resnet101 pretrained: true output_dim: 8 num_downsample: null remove_stride_from_first_conv: false name: maplocnet matching_dim: 8 z_max: 32 x_max: 32 pixel_per_meter: 1 num_scale_bins: 33 num_rotations: 64 map_encoder: embedding_dim: 48 output_dim: 8 num_classes: areas: 7 ways: 10 nodes: 33 backbone: encoder: vgg19 pretrained: false output_scales: - 0 num_downsample: 3 decoder: - 128 - 64 - 64 padding: replicate unary_prior: false bev_net: num_blocks: 4 latent_dim: 128 output_dim: 8 confidence: true experiment: name: maplocanet_523_single_A100_no_mutil_scale_augmentation_resnet101_nosingle gpus: 2 seed: 0 training: lr: 0.0001 lr_scheduler: null finetune_from_checkpoint: null trainer: val_check_interval: 1000 log_every_n_steps: 100 # limit_val_batches: 1000 max_steps: 200000 devices: ${experiment.gpus} checkpointing: monitor: "val/xy_recall_1m" save_top_k: 10 mode: min # filename: '{epoch}-{step}-{loss_SanFrancisco:.2f}'