{
  "model_family": "VisionHOPE",
  "code_repository": "https://github.com/PSRben/VisionHOPE",
  "backbone_defaults": {
    "in_chans": 3,
    "head_dim": 16,
    "chunk_sizes": null,
    "mlp_ratios": [4.0, 4.0, 4.0, 4.0],
    "local_mode": "gated_dwconv7",
    "norm_layer": "layernorm2d",
    "layer_scale_init_value": 1e-06,
    "use_grn": true,
    "use_mlp_dwconv": true,
    "use_direction_scale": true,
    "direction_merge": "sum",
    "use_block_cpe": true,
    "use_output_lepe": true,
    "share_direction_skip": true
  },
  "backbones": {
    "visionhope_tiny": {
      "embed_dims": [64, 128, 256, 512],
      "mixer_dims": [32, 64, 128, 256],
      "depths": [3, 4, 18, 4]
    },
    "visionhope_small": {
      "embed_dims": [64, 160, 320, 512],
      "mixer_dims": [32, 80, 160, 256],
      "depths": [4, 8, 25, 8]
    },
    "visionhope_base": {
      "embed_dims": [96, 192, 448, 640],
      "mixer_dims": [48, 96, 224, 320],
      "depths": [4, 8, 25, 8]
    }
  },
  "tasks": {
    "classification": {
      "dataset": "ImageNet-1K",
      "model_class": "HierarchicalVisionHOPE",
      "model_kwargs": {
        "num_classes": 1000,
        "projection_dim": 1024
      },
      "preprocessing": {
        "input_size": [3, 224, 224],
        "crop_pct": 1.0,
        "interpolation": "bicubic",
        "mean": [0.485, 0.456, 0.406],
        "std": [0.229, 0.224, 0.225]
      },
      "checkpoints": {
        "visionhope_tiny": "visionhope_tiny.pth",
        "visionhope_small": "visionhope_small.pth",
        "visionhope_base": "visionhope_base.pth"
      }
    },
    "detection": {
      "dataset": "COCO",
      "framework": "MMDetection",
      "model_class": "MaskRCNN",
      "num_classes": 80,
      "config_builder": "visionhope.tasks.detection.config.build_config",
      "checkpoints": {
        "1x": {
          "visionhope_tiny": "visionhope_tiny_coco_1x.pth",
          "visionhope_small": "visionhope_small_coco_1x.pth",
          "visionhope_base": "visionhope_base_coco_1x.pth"
        },
        "3x": {
          "visionhope_tiny": "visionhope_tiny_coco_3x.pth",
          "visionhope_small": "visionhope_small_coco_3x.pth"
        }
      }
    },
    "segmentation": {
      "dataset": "ADE20K",
      "framework": "MMSegmentation",
      "model_class": "EncoderDecoder",
      "decode_head": "UPerHead",
      "num_classes": 150,
      "config_builder": "visionhope.tasks.segmentation.config.build_config",
      "checkpoints": {
        "visionhope_tiny": "visionhope_tiny_ade20k.pth",
        "visionhope_small": "visionhope_small_ade20k.pth",
        "visionhope_base": "visionhope_base_ade20k.pth"
      }
    }
  }
}