Evaluation script to reproduce numbers in SAM-HQ paper about sam-hq HOT 4 OPEN

Hi, Thanks for the response. Can you please confirm if the published results can be reproduced using this checkpoint. I tried --restore model from with -restore-model ./pretrained_checkpoint/sam_hq_vit_l.pth and got following errors: RuntimeError: Error(s) in loading state_dict for MaskDecoderHQ: Missing key(s) in state_dict: "transformer.layers.0.self_attn.q_proj.weight", "transformer.layers.0.self_attn.q_proj.bias", "transformer.layers.0.self_attn.k_proj.weight", "transformer.layers.0.self_attn.k_proj.bias", "transformer.layers.0.self_attn.v_proj.weight", "transformer.layers.0.self_attn.v_proj.bias", "transformer.layers.0.self_attn.out_proj.weight", "transformer.layers.0.self_attn.out_proj.bias", "transformer.layers.0.norm1.weight", "transformer.layers.0.norm1.bias", "transformer.layers.0.cross_attn_token_to_image.q_proj.weight", "transformer.layers.0.cross_attn_token_to_image.q_proj.bias", "transformer.layers.0.cross_attn_token_to_image.k_proj.weight", "transformer.layers.0.cross_attn_token_to_image.k_proj.bias", "transformer.layers.0.cross_attn_token_to_image.v_proj.weight", "transformer.layers.0.cross_attn_token_to_image.v_proj.bias", "transformer.layers.0.cross_attn_token_to_image.out_proj.weight", "transformer.layers.0.cross_attn_token_to_image.out_proj.bias", "transformer.layers.0.norm2.weight", "transformer.layers.0.norm2.bias", "transformer.layers.0.mlp.lin1.weight", "transformer.layers.0.mlp.lin1.bias", "transformer.layers.0.mlp.lin2.weight", "transformer.layers.0.mlp.lin2.bias", "transformer.layers.0.norm3.weight", "transformer.layers.0.norm3.bias", "transformer.layers.0.norm4.weight", "transformer.layers.0.norm4.bias", "transformer.layers.0.cross_attn_image_to_token.q_proj.weight", "transformer.layers.0.cross_attn_image_to_token.q_proj.bias", "transformer.layers.0.cross_attn_image_to_token.k_proj.weight", "transformer.layers.0.cross_attn_image_to_token.k_proj.bias", "transformer.layers.0.cross_attn_image_to_token.v_proj.weight", "transformer.layers.0.cross_attn_image_to_token.v_proj.bias", "transformer.layers.0.cross_attn_image_to_token.out_proj.weight", "transformer.layers.0.cross_attn_image_to_token.out_proj.bias", "transformer.layers.1.self_attn.q_proj.weight", "transformer.layers.1.self_attn.q_proj.bias", "transformer.layers.1.self_attn.k_proj.weight", "transformer.layers.1.self_attn.k_proj.bias", "transformer.layers.1.self_attn.v_proj.weight", "transformer.layers.1.self_attn.v_proj.bias", "transformer.layers.1.self_attn.out_proj.weight", "transformer.layers.1.self_attn.out_proj.bias", "transformer.layers.1.norm1.weight", "transformer.layers.1.norm1.bias", "transformer.layers.1.cross_attn_token_to_image.q_proj.weight", "transformer.layers.1.cross_attn_token_to_image.q_proj.bias", "transformer.layers.1.cross_attn_token_to_image.k_proj.weight", "transformer.layers.1.cross_attn_token_to_image.k_proj.bias", "transformer.layers.1.cross_attn_token_to_image.v_proj.weight", "transformer.layers.1.cross_attn_token_to_image.v_proj.bias", "transformer.layers.1.cross_attn_token_to_image.out_proj.weight", "transformer.layers.1.cross_attn_token_to_image.out_proj.bias", "transformer.layers.1.norm2.weight", "transformer.layers.1.norm2.bias", "transformer.layers.1.mlp.lin1.weight", "transformer.layers.1.mlp.lin1.bias", "transformer.layers.1.mlp.lin2.weight", "transformer.layers.1.mlp.lin2.bias", "transformer.layers.1.norm3.weight", "transformer.layers.1.norm3.bias", "transformer.layers.1.norm4.weight", "transformer.layers.1.norm4.bias", "transformer.layers.1.cross_attn_image_to_token.q_proj.weight", "transformer.layers.1.cross_attn_image_to_token.q_proj.bias", "transformer.layers.1.cross_attn_image_to_token.k_proj.weight", "transformer.layers.1.cross_attn_image_to_token.k_proj.bias", "transformer.layers.1.cross_attn_image_to_token.v_proj.weight", "transformer.layers.1.cross_attn_image_to_token.v_proj.bias", "transformer.layers.1.cross_attn_image_to_token.out_proj.weight", "transformer.layers.1.cross_attn_image_to_token.out_proj.bias", "transformer.final_attn_token_to_image.q_proj.weight", "transformer.final_attn_token_to_image.q_proj.bias", "transformer.final_attn_token_to_image.k_proj.weight", "transformer.final_attn_token_to_image.k_proj.bias", "transformer.final_attn_token_to_image.v_proj.weight", "transformer.final_attn_token_to_image.v_proj.bias", "transformer.final_attn_token_to_image.out_proj.weight", "transformer.final_attn_token_to_image.out_proj.bias", "transformer.norm_final_attn.weight", "transformer.norm_final_attn.bias", "iou_token.weight", "mask_tokens.weight", "output_upscaling.0.weight", "output_upscaling.0.bias", "output_upscaling.1.weight", "output_upscaling.1.bias", "output_upscaling.3.weight", "output_upscaling.3.bias", "output_hypernetworks_mlps.0.layers.0.weight", "output_hypernetworks_mlps.0.layers.0.bias", "output_hypernetworks_mlps.0.layers.1.weight", "output_hypernetworks_mlps.0.layers.1.bias", "output_hypernetworks_mlps.0.layers.2.weight", "output_hypernetworks_mlps.0.layers.2.bias", "output_hypernetworks_mlps.1.layers.0.weight", "output_hypernetworks_mlps.1.layers.0.bias", "output_hypernetworks_mlps.1.layers.1.weight", "output_hypernetworks_mlps.1.layers.1.bias", "output_hypernetworks_mlps.1.layers.2.weight", "output_hypernetworks_mlps.1.layers.2.bias", "output_hypernetworks_mlps.2.layers.0.weight", "output_hypernetworks_mlps.2.layers.0.bias", "output_hypernetworks_mlps.2.layers.1.weight", "output_hypernetworks_mlps.2.layers.1.bias", "output_hypernetworks_mlps.2.layers.2.weight", "output_hypernetworks_mlps.2.layers.2.bias", "output_hypernetworks_mlps.3.layers.0.weight", "output_hypernetworks_mlps.3.layers.0.bias", "output_hypernetworks_mlps.3.layers.1.weight", "output_hypernetworks_mlps.3.layers.1.bias", "output_hypernetworks_mlps.3.layers.2.weight", "output_hypernetworks_mlps.3.layers.2.bias", "iou_prediction_head.layers.0.weight", "iou_prediction_head.layers.0.bias", "iou_prediction_head.layers.1.weight", "iou_prediction_head.layers.1.bias", "iou_prediction_head.layers.2.weight", "iou_prediction_head.layers.2.bias", "hf_token.weight", "hf_mlp.layers.0.weight", "hf_mlp.layers.0.bias", "hf_mlp.layers.1.weight", "hf_mlp.layers.1.bias", "hf_mlp.layers.2.weight", "hf_mlp.layers.2.bias", "compress_vit_feat.0.weight", "compress_vit_feat.0.bias", "compress_vit_feat.1.weight", "compress_vit_feat.1.bias", "compress_vit_feat.3.weight", "compress_vit_feat.3.bias", "embedding_encoder.0.weight", "embedding_encoder.0.bias", "embedding_encoder.1.weight", "embedding_encoder.1.bias", "embedding_encoder.3.weight", "embedding_encoder.3.bias", "embedding_maskfeature.0.weight", "embedding_maskfeature.0.bias", "embedding_maskfeature.1.weight", "embedding_maskfeature.1.bias", "embedding_maskfeature.3.weight", "embedding_maskfeature.3.bias". Unexpected key(s) in state_dict: "image_encoder.neck.0.weight", "image_encoder.neck.1.weight", "image_encoder.neck.1.bias", "image_encoder.neck.2.weight", "image_encoder.neck.3.weight", "image_encoder.neck.3.bias", "image_encoder.patch_embed.proj.weight", "image_encoder.patch_embed.proj.bias", "image_encoder.blocks.0.norm1.weight", "image_encoder.blocks.0.norm1.bias", "image_encoder.blocks.0.attn.rel_pos_h", "image_encoder.blocks.0.attn.rel_pos_w", "image_encoder.blocks.0.attn.qkv.weight", "image_encod Please advise, which checkpoint to use to reproduce results. Thanks, Ankit