configs/config_pop909_256x256_chords.json

{
    "model": {
        "type": "image_transformer_v2",
        "input_channels": 3,
        "input_size": [256, 256],
        "patch_size": [4, 4],
        "depths": [2, 2, 4],
        "widths": [128, 256, 512],
        "self_attns": [
            {"type": "neighborhood", "d_head": 64, "kernel_size": 7},
            {"type": "neighborhood", "d_head": 64, "kernel_size": 7},
            {"type": "global", "d_head": 64}
        ],
        "loss_config": "karras",
        "loss_weighting": "soft-min-snr",
        "dropout_rate": [0.0, 0.0, 0.1],
        "mapping_dropout_rate": 0.0,
        "augment_prob": 0.0,
        "vert_trans_prob": 0.0,
        "sigma_data": 0.5,
        "sigma_min": 1e-2,
        "sigma_max": 160,
        "sigma_sample_density": {
            "type": "cosine-interpolated"
        }
    },
    "dataset": {
        "type": "midi_pianoroll",
        "location": "/data/POP909-Dataset/images_128_rg_chords_TOTAL",
        "crop_size": [128, 512]
    },
    "optimizer": {
        "type": "adamw",
        "lr": 5e-4,
        "betas": [0.9, 0.95],
        "eps": 1e-8,
        "weight_decay": 1e-3
    },
    "lr_sched": {
        "type": "constant",
        "warmup": 0.0
    },
    "ema_sched": {
        "type": "inverse",
        "power": 0.75,
        "max_value": 0.9999
    }
}