{ "aggregate": "semantic", "default_prompt": "the main foreground subject", "detr_decoder_num_layers": 6, "detr_encoder_num_layers": 6, "fpn_hidden_size": 256, "geometry_num_layers": 3, "hidden_size": 256, "image_size": 1008, "intermediate_size": 2048, "nobg_version": "0.3.1", "num_attention_heads": 8, "num_queries": 200, "num_upsampling_stages": 3, "score_threshold": 0.3, "text_hidden_size": 1024, "text_intermediate_size": 4096, "text_max_position_embeddings": 32, "text_num_attention_heads": 16, "text_num_hidden_layers": 24, "text_projection_dim": 512, "text_vocab_size": 49408, "vision_global_attn_indexes": [ 7, 15, 23, 31 ], "vision_hidden_size": 1024, "vision_intermediate_size": 4736, "vision_num_attention_heads": 16, "vision_num_hidden_layers": 32, "vision_patch_size": 14, "vision_pretrain_image_size": 336, "vision_window_size": 24 }