config.json

{
  "audio_drop_path": 0.1,
  "audio_embed_dim": 768,
  "audio_kernel_size": 16,
  "audio_num_blocks": 12,
  "audio_num_heads": 12,
  "audio_num_mel_bins": 128,
  "audio_stride": 10,
  "audio_target_len": 204,
  "depth_drop_path": 0.0,
  "depth_embed_dim": 384,
  "depth_kernel_size": 16,
  "depth_num_blocks": 12,
  "depth_num_heads": 8,
  "imu_drop_path": 0.7,
  "imu_embed_dim": 512,
  "imu_kernel_size": 8,
  "imu_num_blocks": 6,
  "imu_num_heads": 8,
  "kernel_size": [
    2,
    14,
    14
  ],
  "out_embed_dim": 1024,
  "text_embed_dim": 1024,
  "text_num_blocks": 24,
  "text_num_heads": 16,
  "thermal_drop_path": 0.0,
  "thermal_embed_dim": 768,
  "thermal_kernel_size": 16,
  "thermal_num_blocks": 12,
  "thermal_num_heads": 12,
  "video_frames": 2,
  "vision_embed_dim": 1280,
  "vision_num_blocks": 32,
  "vision_num_heads": 16
}