config.json

{
    "backbone_config": {
        "model_type": "sapiens2_backbone",
        "arch": "sapiens2_0.4b",
        "embed_dims": 1024,
        "num_layers": 24,
        "num_heads": 16,
        "feedforward_channels": 4096,
        "image_size": [
            1024,
            768
        ],
        "patch_size": 16,
        "n_storage_tokens": 8,
        "rope_base": 100.0,
        "rope_normalize_coords": "separate",
        "mhsa_early": 8,
        "mhsa_late": 8,
        "layer_scale_init_value": 0.0001,
        "final_norm": true
    },
    "head_config": {
        "model_type": "sapiens2_head",
        "task": "seg",
        "in_channels": 1024,
        "deconv_out_channels": [
            512,
            256,
            128,
            64
        ],
        "deconv_kernel_sizes": [
            4,
            4,
            4,
            4
        ],
        "conv_out_channels": [
            64,
            64
        ],
        "conv_kernel_sizes": [
            1,
            1
        ],
        "num_keypoints": 308,
        "num_classes": 29,
        "upsample_channels": null,
        "scale_conv_out_channels": null,
        "scale_conv_kernel_sizes": null,
        "scale_final_layer": null
    },
    "model_type": "sapiens2",
    "quantization": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine"
    },
    "quantization_config": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine"
    },
    "size": "0.4b",
    "task": "seg",
    "vision_config": {}
}