Mr.DETR++预训练权重加载问题

[ffxxy]

WARNING [08/08 09:44:03 fvcore.common.checkpoint]: Some model parameters or buffers are not found in the checkpoint:
backbone.net.blocks.0.attn.proj.{bias, weight}
backbone.net.blocks.0.attn.qkv.{bias, weight}
backbone.net.blocks.0.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.0.mlp.fc1.{bias, weight}
backbone.net.blocks.0.mlp.fc2.{bias, weight}
backbone.net.blocks.0.norm1.{bias, weight}
backbone.net.blocks.0.norm2.{bias, weight}
backbone.net.blocks.1.attn.proj.{bias, weight}
backbone.net.blocks.1.attn.qkv.{bias, weight}
backbone.net.blocks.1.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.1.mlp.fc1.{bias, weight}
backbone.net.blocks.1.mlp.fc2.{bias, weight}
backbone.net.blocks.1.norm1.{bias, weight}
backbone.net.blocks.1.norm2.{bias, weight}
backbone.net.blocks.10.attn.proj.{bias, weight}
backbone.net.blocks.10.attn.qkv.{bias, weight}
backbone.net.blocks.10.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.10.mlp.fc1.{bias, weight}
backbone.net.blocks.10.mlp.fc2.{bias, weight}
backbone.net.blocks.10.norm1.{bias, weight}
backbone.net.blocks.10.norm2.{bias, weight}
backbone.net.blocks.11.attn.proj.{bias, weight}
backbone.net.blocks.11.attn.qkv.{bias, weight}
backbone.net.blocks.11.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.11.mlp.fc1.{bias, weight}
backbone.net.blocks.11.mlp.fc2.{bias, weight}
backbone.net.blocks.11.norm1.{bias, weight}
backbone.net.blocks.11.norm2.{bias, weight}
backbone.net.blocks.12.attn.proj.{bias, weight}
backbone.net.blocks.12.attn.qkv.{bias, weight}
backbone.net.blocks.12.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.12.mlp.fc1.{bias, weight}
backbone.net.blocks.12.mlp.fc2.{bias, weight}
backbone.net.blocks.12.norm1.{bias, weight}
backbone.net.blocks.12.norm2.{bias, weight}
backbone.net.blocks.13.attn.proj.{bias, weight}
backbone.net.blocks.13.attn.qkv.{bias, weight}
backbone.net.blocks.13.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.13.mlp.fc1.{bias, weight}
backbone.net.blocks.13.mlp.fc2.{bias, weight}
backbone.net.blocks.13.norm1.{bias, weight}
backbone.net.blocks.13.norm2.{bias, weight}
backbone.net.blocks.14.attn.proj.{bias, weight}
backbone.net.blocks.14.attn.qkv.{bias, weight}
backbone.net.blocks.14.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.14.mlp.fc1.{bias, weight}
backbone.net.blocks.14.mlp.fc2.{bias, weight}
backbone.net.blocks.14.norm1.{bias, weight}
backbone.net.blocks.14.norm2.{bias, weight}
backbone.net.blocks.15.attn.proj.{bias, weight}
backbone.net.blocks.15.attn.qkv.{bias, weight}
backbone.net.blocks.15.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.15.mlp.fc1.{bias, weight}
backbone.net.blocks.15.mlp.fc2.{bias, weight}
backbone.net.blocks.15.norm1.{bias, weight}
backbone.net.blocks.15.norm2.{bias, weight}
backbone.net.blocks.16.attn.proj.{bias, weight}
backbone.net.blocks.16.attn.qkv.{bias, weight}
backbone.net.blocks.16.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.16.mlp.fc1.{bias, weight}
backbone.net.blocks.16.mlp.fc2.{bias, weight}
backbone.net.blocks.16.norm1.{bias, weight}
backbone.net.blocks.16.norm2.{bias, weight}
backbone.net.blocks.17.attn.proj.{bias, weight}
backbone.net.blocks.17.attn.qkv.{bias, weight}
backbone.net.blocks.17.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.17.mlp.fc1.{bias, weight}
backbone.net.blocks.17.mlp.fc2.{bias, weight}
backbone.net.blocks.17.norm1.{bias, weight}
backbone.net.blocks.17.norm2.{bias, weight}
backbone.net.blocks.18.attn.proj.{bias, weight}
backbone.net.blocks.18.attn.qkv.{bias, weight}
backbone.net.blocks.18.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.18.mlp.fc1.{bias, weight}
backbone.net.blocks.18.mlp.fc2.{bias, weight}
backbone.net.blocks.18.norm1.{bias, weight}
backbone.net.blocks.18.norm2.{bias, weight}
backbone.net.blocks.19.attn.proj.{bias, weight}
backbone.net.blocks.19.attn.qkv.{bias, weight}
backbone.net.blocks.19.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.19.mlp.fc1.{bias, weight}
backbone.net.blocks.19.mlp.fc2.{bias, weight}
backbone.net.blocks.19.norm1.{bias, weight}
backbone.net.blocks.19.norm2.{bias, weight}
backbone.net.blocks.2.attn.proj.{bias, weight}
backbone.net.blocks.2.attn.qkv.{bias, weight}
backbone.net.blocks.2.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.2.mlp.fc1.{bias, weight}
backbone.net.blocks.2.mlp.fc2.{bias, weight}
backbone.net.blocks.2.norm1.{bias, weight}
backbone.net.blocks.2.norm2.{bias, weight}
backbone.net.blocks.20.attn.proj.{bias, weight}
backbone.net.blocks.20.attn.qkv.{bias, weight}
backbone.net.blocks.20.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.20.mlp.fc1.{bias, weight}
backbone.net.blocks.20.mlp.fc2.{bias, weight}
backbone.net.blocks.20.norm1.{bias, weight}
backbone.net.blocks.20.norm2.{bias, weight}
backbone.net.blocks.21.attn.proj.{bias, weight}
backbone.net.blocks.21.attn.qkv.{bias, weight}
backbone.net.blocks.21.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.21.mlp.fc1.{bias, weight}
backbone.net.blocks.21.mlp.fc2.{bias, weight}
backbone.net.blocks.21.norm1.{bias, weight}
backbone.net.blocks.21.norm2.{bias, weight}
backbone.net.blocks.22.attn.proj.{bias, weight}
backbone.net.blocks.22.attn.qkv.{bias, weight}
backbone.net.blocks.22.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.22.mlp.fc1.{bias, weight}
backbone.net.blocks.22.mlp.fc2.{bias, weight}
backbone.net.blocks.22.norm1.{bias, weight}
backbone.net.blocks.22.norm2.{bias, weight}
backbone.net.blocks.23.attn.proj.{bias, weight}
backbone.net.blocks.23.attn.qkv.{bias, weight}
backbone.net.blocks.23.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.23.mlp.fc1.{bias, weight}
backbone.net.blocks.23.mlp.fc2.{bias, weight}
backbone.net.blocks.23.norm1.{bias, weight}
backbone.net.blocks.23.norm2.{bias, weight}
backbone.net.blocks.3.attn.proj.{bias, weight}
backbone.net.blocks.3.attn.qkv.{bias, weight}
backbone.net.blocks.3.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.3.mlp.fc1.{bias, weight}
backbone.net.blocks.3.mlp.fc2.{bias, weight}
backbone.net.blocks.3.norm1.{bias, weight}
backbone.net.blocks.3.norm2.{bias, weight}
backbone.net.blocks.4.attn.proj.{bias, weight}
backbone.net.blocks.4.attn.qkv.{bias, weight}
backbone.net.blocks.4.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.4.mlp.fc1.{bias, weight}
backbone.net.blocks.4.mlp.fc2.{bias, weight}
backbone.net.blocks.4.norm1.{bias, weight}
backbone.net.blocks.4.norm2.{bias, weight}
backbone.net.blocks.5.attn.proj.{bias, weight}
backbone.net.blocks.5.attn.qkv.{bias, weight}
backbone.net.blocks.5.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.5.mlp.fc1.{bias, weight}
backbone.net.blocks.5.mlp.fc2.{bias, weight}
backbone.net.blocks.5.norm1.{bias, weight}
backbone.net.blocks.5.norm2.{bias, weight}
backbone.net.blocks.6.attn.proj.{bias, weight}
backbone.net.blocks.6.attn.qkv.{bias, weight}
backbone.net.blocks.6.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.6.mlp.fc1.{bias, weight}
backbone.net.blocks.6.mlp.fc2.{bias, weight}
backbone.net.blocks.6.norm1.{bias, weight}
backbone.net.blocks.6.norm2.{bias, weight}
backbone.net.blocks.7.attn.proj.{bias, weight}
backbone.net.blocks.7.attn.qkv.{bias, weight}
backbone.net.blocks.7.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.7.mlp.fc1.{bias, weight}
backbone.net.blocks.7.mlp.fc2.{bias, weight}
backbone.net.blocks.7.norm1.{bias, weight}
backbone.net.blocks.7.norm2.{bias, weight}
backbone.net.blocks.8.attn.proj.{bias, weight}
backbone.net.blocks.8.attn.qkv.{bias, weight}
backbone.net.blocks.8.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.8.mlp.fc1.{bias, weight}
backbone.net.blocks.8.mlp.fc2.{bias, weight}
backbone.net.blocks.8.norm1.{bias, weight}
backbone.net.blocks.8.norm2.{bias, weight}
backbone.net.blocks.9.attn.proj.{bias, weight}
backbone.net.blocks.9.attn.qkv.{bias, weight}
backbone.net.blocks.9.attn.{rel_pos_h, rel_pos_w}
backbone.net.blocks.9.mlp.fc1.{bias, weight}
backbone.net.blocks.9.mlp.fc2.{bias, weight}
backbone.net.blocks.9.norm1.{bias, weight}
backbone.net.blocks.9.norm2.{bias, weight}
backbone.net.patch_embed.proj.{bias, weight}
backbone.net.pos_embed
backbone.simfp_3.0.{bias, weight}
backbone.simfp_3.1.norm.{bias, weight}
backbone.simfp_3.1.weight
backbone.simfp_3.2.norm.{bias, weight}
backbone.simfp_3.2.weight
backbone.simfp_4.0.norm.{bias, weight}
backbone.simfp_4.0.weight
backbone.simfp_4.1.norm.{bias, weight}
backbone.simfp_4.1.weight
backbone.simfp_5.1.norm.{bias, weight}
backbone.simfp_5.1.weight
backbone.simfp_5.2.norm.{bias, weight}
backbone.simfp_5.2.weight
bbox_embed.0.layers.0.{bias, weight}
bbox_embed.0.layers.1.{bias, weight}
bbox_embed.0.layers.2.{bias, weight}
bbox_embed.1.layers.0.{bias, weight}
bbox_embed.1.layers.1.{bias, weight}
bbox_embed.1.layers.2.{bias, weight}
bbox_embed.2.layers.0.{bias, weight}
bbox_embed.2.layers.1.{bias, weight}
bbox_embed.2.layers.2.{bias, weight}
bbox_embed.3.layers.0.{bias, weight}
bbox_embed.3.layers.1.{bias, weight}
bbox_embed.3.layers.2.{bias, weight}
bbox_embed.4.layers.0.{bias, weight}
bbox_embed.4.layers.1.{bias, weight}
bbox_embed.4.layers.2.{bias, weight}
bbox_embed.5.layers.0.{bias, weight}
bbox_embed.5.layers.1.{bias, weight}
bbox_embed.5.layers.2.{bias, weight}
bbox_embed.6.layers.0.{bias, weight}
bbox_embed.6.layers.1.{bias, weight}
bbox_embed.6.layers.2.{bias, weight}
class_embed.0.{bias, weight}
class_embed.1.{bias, weight}
class_embed.2.{bias, weight}
class_embed.3.{bias, weight}
class_embed.4.{bias, weight}
class_embed.5.{bias, weight}
class_embed.6.{bias, weight}
label_enc.weight
neck.convs.0.conv.{bias, weight}
neck.convs.0.norm.{bias, weight}
neck.convs.1.conv.{bias, weight}
neck.convs.1.norm.{bias, weight}
neck.convs.2.conv.{bias, weight}
neck.convs.2.norm.{bias, weight}
neck.convs.3.conv.{bias, weight}
neck.convs.3.norm.{bias, weight}
transformer.decoder.bbox_embed.0.layers.0.{bias, weight}
transformer.decoder.bbox_embed.0.layers.1.{bias, weight}
transformer.decoder.bbox_embed.0.layers.2.{bias, weight}
transformer.decoder.bbox_embed.1.layers.0.{bias, weight}
transformer.decoder.bbox_embed.1.layers.1.{bias, weight}
transformer.decoder.bbox_embed.1.layers.2.{bias, weight}
transformer.decoder.bbox_embed.2.layers.0.{bias, weight}
transformer.decoder.bbox_embed.2.layers.1.{bias, weight}
transformer.decoder.bbox_embed.2.layers.2.{bias, weight}
transformer.decoder.bbox_embed.3.layers.0.{bias, weight}
transformer.decoder.bbox_embed.3.layers.1.{bias, weight}
transformer.decoder.bbox_embed.3.layers.2.{bias, weight}
transformer.decoder.bbox_embed.4.layers.0.{bias, weight}
transformer.decoder.bbox_embed.4.layers.1.{bias, weight}
transformer.decoder.bbox_embed.4.layers.2.{bias, weight}
transformer.decoder.bbox_embed.5.layers.0.{bias, weight}
transformer.decoder.bbox_embed.5.layers.1.{bias, weight}
transformer.decoder.bbox_embed.5.layers.2.{bias, weight}
transformer.decoder.bbox_embed.6.layers.0.{bias, weight}
transformer.decoder.bbox_embed.6.layers.1.{bias, weight}
transformer.decoder.bbox_embed.6.layers.2.{bias, weight}
transformer.decoder.bbox_embed_o2m.layers.0.{bias, weight}
transformer.decoder.bbox_embed_o2m.layers.1.{bias, weight}
transformer.decoder.bbox_embed_o2m.layers.2.{bias, weight}
transformer.decoder.class_embed.0.{bias, weight}
transformer.decoder.class_embed.1.{bias, weight}
transformer.decoder.class_embed.2.{bias, weight}
transformer.decoder.class_embed.3.{bias, weight}
transformer.decoder.class_embed.4.{bias, weight}
transformer.decoder.class_embed.5.{bias, weight}
transformer.decoder.class_embed.6.{bias, weight}
transformer.decoder.class_embed_iou.0.{bias, weight}
transformer.decoder.class_embed_iou.1.{bias, weight}
transformer.decoder.class_embed_iou.2.{bias, weight}
transformer.decoder.class_embed_iou.3.{bias, weight}
transformer.decoder.class_embed_iou.4.{bias, weight}
transformer.decoder.class_embed_iou.5.{bias, weight}
transformer.decoder.class_embed_iou.6.{bias, weight}
transformer.decoder.class_embed_o2m.{bias, weight}
transformer.decoder.layers.0.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.0.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.0.attentions.0.instruction_o2m.weight
transformer.decoder.layers.0.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.0.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.0.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.0.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.0.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.0.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.0.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.0.ffns.0.layers.0.linear.linear.{bias, weight}
transformer.decoder.layers.0.ffns.0.layers.0.linear.linear_sep.{bias, weight}
transformer.decoder.layers.0.ffns.0.layers.1.linear.{bias, weight}
transformer.decoder.layers.0.ffns.0.layers.1.linear_sep.{bias, weight}
transformer.decoder.layers.0.norms.0.{bias, weight}
transformer.decoder.layers.0.norms.1.{bias, weight}
transformer.decoder.layers.1.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.1.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.1.attentions.0.instruction_o2m.weight
transformer.decoder.layers.1.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.1.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.1.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.1.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.1.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.1.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.1.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.1.ffns.0.layers.0.linear.linear.{bias, weight}
transformer.decoder.layers.1.ffns.0.layers.0.linear.linear_sep.{bias, weight}
transformer.decoder.layers.1.ffns.0.layers.1.linear.{bias, weight}
transformer.decoder.layers.1.ffns.0.layers.1.linear_sep.{bias, weight}
transformer.decoder.layers.1.norms.0.{bias, weight}
transformer.decoder.layers.1.norms.1.{bias, weight}
transformer.decoder.layers.2.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.2.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.2.attentions.0.instruction_o2m.weight
transformer.decoder.layers.2.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.2.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.2.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.2.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.2.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.2.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.2.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.2.ffns.0.layers.0.linear.linear.{bias, weight}
transformer.decoder.layers.2.ffns.0.layers.0.linear.linear_sep.{bias, weight}
transformer.decoder.layers.2.ffns.0.layers.1.linear.{bias, weight}
transformer.decoder.layers.2.ffns.0.layers.1.linear_sep.{bias, weight}
transformer.decoder.layers.2.norms.0.{bias, weight}
transformer.decoder.layers.2.norms.1.{bias, weight}
transformer.decoder.layers.3.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.3.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.3.attentions.0.instruction_o2m.weight
transformer.decoder.layers.3.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.3.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.3.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.3.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.3.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.3.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.3.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.3.ffns.0.layers.0.linear.linear.{bias, weight}
transformer.decoder.layers.3.ffns.0.layers.0.linear.linear_sep.{bias, weight}
transformer.decoder.layers.3.ffns.0.layers.1.linear.{bias, weight}
transformer.decoder.layers.3.ffns.0.layers.1.linear_sep.{bias, weight}
transformer.decoder.layers.3.norms.0.{bias, weight}
transformer.decoder.layers.3.norms.1.{bias, weight}
transformer.decoder.layers.4.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.4.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.4.attentions.0.instruction_o2m.weight
transformer.decoder.layers.4.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.4.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.4.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.4.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.4.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.4.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.4.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.4.ffns.0.layers.0.linear.linear.{bias, weight}
transformer.decoder.layers.4.ffns.0.layers.0.linear.linear_sep.{bias, weight}
transformer.decoder.layers.4.ffns.0.layers.1.linear.{bias, weight}
transformer.decoder.layers.4.ffns.0.layers.1.linear_sep.{bias, weight}
transformer.decoder.layers.4.norms.0.{bias, weight}
transformer.decoder.layers.4.norms.1.{bias, weight}
transformer.decoder.layers.5.attentions.0.attn.out_proj.{bias, weight}
transformer.decoder.layers.5.attentions.0.attn.{in_proj_bias, in_proj_weight}
transformer.decoder.layers.5.attentions.0.instruction_o2m.weight
transformer.decoder.layers.5.attentions.0.instruction_o2m_pos.weight
transformer.decoder.layers.5.attentions.0.norm_o2m.{bias, weight}
transformer.decoder.layers.5.attentions.0.norm_o2o.{bias, weight}
transformer.decoder.layers.5.attentions.1.attention_weights.{bias, weight}
transformer.decoder.layers.5.attentions.1.output_proj.{bias, weight}
transformer.decoder.layers.5.attentions.1.sampling_offsets.{bias, weight}
transformer.decoder.layers.5.attentions.1.value_proj.{bias, weight}
transformer.decoder.layers.5.ffns.0.experts.0.down_proj.weight
transformer.decoder.layers.5.ffns.0.experts.0.gate_proj.weight
transformer.decoder.layers.5.ffns.0.experts.0.up_proj.weight
transformer.decoder.layers.5.ffns.0.experts.1.down_proj.weight
transformer.decoder.layers.5.ffns.0.experts.1.gate_proj.weight
transformer.decoder.layers.5.ffns.0.experts.1.up_proj.weight
transformer.decoder.layers.5.ffns.0.experts.2.down_proj.weight
transformer.decoder.layers.5.ffns.0.experts.2.gate_proj.weight
transformer.decoder.layers.5.ffns.0.experts.2.up_proj.weight
transformer.decoder.layers.5.ffns.0.experts.3.down_proj.weight
transformer.decoder.layers.5.ffns.0.experts.3.gate_proj.weight
transformer.decoder.layers.5.ffns.0.experts.3.up_proj.weight
transformer.decoder.layers.5.ffns.0.gate.weight
transformer.decoder.layers.5.ffns.0.gate_sep.weight
transformer.decoder.layers.5.norms.0.{bias, weight}
transformer.decoder.layers.5.norms.1.{bias, weight}
transformer.decoder.norm.{bias, weight}
transformer.decoder.ref_point_head.layers.0.{bias, weight}
transformer.decoder.ref_point_head.layers.1.{bias, weight}
transformer.enc_output.{bias, weight}
transformer.enc_output_norm.{bias, weight}
transformer.encoder.layers.0.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.0.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.0.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.0.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.0.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.0.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.0.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.0.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.0.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.0.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.0.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.0.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.0.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.0.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.0.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.0.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.0.ffns.0.gate.weight
transformer.encoder.layers.0.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.0.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.0.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.0.norms.0.{bias, weight}
transformer.encoder.layers.0.norms.1.{bias, weight}
transformer.encoder.layers.1.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.1.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.1.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.1.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.1.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.1.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.1.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.1.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.1.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.1.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.1.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.1.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.1.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.1.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.1.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.1.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.1.ffns.0.gate.weight
transformer.encoder.layers.1.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.1.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.1.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.1.norms.0.{bias, weight}
transformer.encoder.layers.1.norms.1.{bias, weight}
transformer.encoder.layers.2.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.2.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.2.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.2.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.2.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.2.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.2.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.2.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.2.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.2.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.2.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.2.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.2.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.2.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.2.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.2.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.2.ffns.0.gate.weight
transformer.encoder.layers.2.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.2.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.2.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.2.norms.0.{bias, weight}
transformer.encoder.layers.2.norms.1.{bias, weight}
transformer.encoder.layers.3.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.3.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.3.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.3.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.3.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.3.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.3.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.3.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.3.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.3.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.3.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.3.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.3.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.3.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.3.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.3.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.3.ffns.0.gate.weight
transformer.encoder.layers.3.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.3.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.3.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.3.norms.0.{bias, weight}
transformer.encoder.layers.3.norms.1.{bias, weight}
transformer.encoder.layers.4.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.4.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.4.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.4.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.4.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.4.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.4.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.4.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.4.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.4.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.4.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.4.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.4.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.4.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.4.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.4.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.4.ffns.0.gate.weight
transformer.encoder.layers.4.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.4.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.4.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.4.norms.0.{bias, weight}
transformer.encoder.layers.4.norms.1.{bias, weight}
transformer.encoder.layers.5.attentions.0.attention_weights.{bias, weight}
transformer.encoder.layers.5.attentions.0.output_proj.{bias, weight}
transformer.encoder.layers.5.attentions.0.sampling_offsets.{bias, weight}
transformer.encoder.layers.5.attentions.0.value_proj.{bias, weight}
transformer.encoder.layers.5.ffns.0.experts.0.down_proj.weight
transformer.encoder.layers.5.ffns.0.experts.0.gate_proj.weight
transformer.encoder.layers.5.ffns.0.experts.0.up_proj.weight
transformer.encoder.layers.5.ffns.0.experts.1.down_proj.weight
transformer.encoder.layers.5.ffns.0.experts.1.gate_proj.weight
transformer.encoder.layers.5.ffns.0.experts.1.up_proj.weight
transformer.encoder.layers.5.ffns.0.experts.2.down_proj.weight
transformer.encoder.layers.5.ffns.0.experts.2.gate_proj.weight
transformer.encoder.layers.5.ffns.0.experts.2.up_proj.weight
transformer.encoder.layers.5.ffns.0.experts.3.down_proj.weight
transformer.encoder.layers.5.ffns.0.experts.3.gate_proj.weight
transformer.encoder.layers.5.ffns.0.experts.3.up_proj.weight
transformer.encoder.layers.5.ffns.0.gate.weight
transformer.encoder.layers.5.ffns.0.shared_experts.down_proj.weight
transformer.encoder.layers.5.ffns.0.shared_experts.gate_proj.weight
transformer.encoder.layers.5.ffns.0.shared_experts.up_proj.weight
transformer.encoder.layers.5.norms.0.{bias, weight}
transformer.encoder.layers.5.norms.1.{bias, weight}
transformer.encoder.mapping_func.0.0.linear.{bias, weight}
transformer.encoder.mapping_func.0.1.{bias, weight}
transformer.encoder.mapping_func.1.0.linear.{bias, weight}
transformer.encoder.mapping_func.1.1.{bias, weight}
transformer.encoder.mapping_func.2.0.linear.{bias, weight}
transformer.encoder.mapping_func.2.1.{bias, weight}
transformer.encoder.mapping_func.3.0.linear.{bias, weight}
transformer.encoder.mapping_func.3.1.{bias, weight}
transformer.encoder.mapping_func.4.0.linear.{bias, weight}
transformer.encoder.mapping_func.4.1.{bias, weight}
transformer.encoder.mapping_func.5.0.linear.{bias, weight}
transformer.encoder.mapping_func.5.1.{bias, weight}
transformer.encoder.mapping_func.6.0.linear.{bias, weight}
transformer.encoder.mapping_func.6.1.{bias, weight}
transformer.level_embeds
transformer.tgt_embed.weight
WARNING [08/08 09:44:03 fvcore.common.checkpoint]: The checkpoint state_dict contains keys that are not used by the model:
cls_token
pos_embed
patch_embed.proj.{bias, weight}
blocks.0.norm1.{bias, weight}
blocks.0.attn.qkv.{bias, weight}
blocks.0.attn.proj.{bias, weight}
blocks.0.norm2.{bias, weight}
blocks.0.mlp.fc1.{bias, weight}
blocks.0.mlp.fc2.{bias, weight}
blocks.1.norm1.{bias, weight}
blocks.1.attn.qkv.{bias, weight}
blocks.1.attn.proj.{bias, weight}
blocks.1.norm2.{bias, weight}
blocks.1.mlp.fc1.{bias, weight}
blocks.1.mlp.fc2.{bias, weight}
blocks.2.norm1.{bias, weight}
blocks.2.attn.qkv.{bias, weight}
blocks.2.attn.proj.{bias, weight}
blocks.2.norm2.{bias, weight}
blocks.2.mlp.fc1.{bias, weight}
blocks.2.mlp.fc2.{bias, weight}
blocks.3.norm1.{bias, weight}
blocks.3.attn.qkv.{bias, weight}
blocks.3.attn.proj.{bias, weight}
blocks.3.norm2.{bias, weight}
blocks.3.mlp.fc1.{bias, weight}
blocks.3.mlp.fc2.{bias, weight}
blocks.4.norm1.{bias, weight}
blocks.4.attn.qkv.{bias, weight}
blocks.4.attn.proj.{bias, weight}
blocks.4.norm2.{bias, weight}
blocks.4.mlp.fc1.{bias, weight}
blocks.4.mlp.fc2.{bias, weight}
blocks.5.norm1.{bias, weight}
blocks.5.attn.qkv.{bias, weight}
blocks.5.attn.proj.{bias, weight}
blocks.5.norm2.{bias, weight}
blocks.5.mlp.fc1.{bias, weight}
blocks.5.mlp.fc2.{bias, weight}
blocks.6.norm1.{bias, weight}
blocks.6.attn.qkv.{bias, weight}
blocks.6.attn.proj.{bias, weight}
blocks.6.norm2.{bias, weight}
blocks.6.mlp.fc1.{bias, weight}
blocks.6.mlp.fc2.{bias, weight}
blocks.7.norm1.{bias, weight}
blocks.7.attn.qkv.{bias, weight}
blocks.7.attn.proj.{bias, weight}
blocks.7.norm2.{bias, weight}
blocks.7.mlp.fc1.{bias, weight}
blocks.7.mlp.fc2.{bias, weight}
blocks.8.norm1.{bias, weight}
blocks.8.attn.qkv.{bias, weight}
blocks.8.attn.proj.{bias, weight}
blocks.8.norm2.{bias, weight}
blocks.8.mlp.fc1.{bias, weight}
blocks.8.mlp.fc2.{bias, weight}
blocks.9.norm1.{bias, weight}
blocks.9.attn.qkv.{bias, weight}
blocks.9.attn.proj.{bias, weight}
blocks.9.norm2.{bias, weight}
blocks.9.mlp.fc1.{bias, weight}
blocks.9.mlp.fc2.{bias, weight}
blocks.10.norm1.{bias, weight}
blocks.10.attn.qkv.{bias, weight}
blocks.10.attn.proj.{bias, weight}
blocks.10.norm2.{bias, weight}
blocks.10.mlp.fc1.{bias, weight}
blocks.10.mlp.fc2.{bias, weight}
blocks.11.norm1.{bias, weight}
blocks.11.attn.qkv.{bias, weight}
blocks.11.attn.proj.{bias, weight}
blocks.11.norm2.{bias, weight}
blocks.11.mlp.fc1.{bias, weight}
blocks.11.mlp.fc2.{bias, weight}
blocks.12.norm1.{bias, weight}
blocks.12.attn.qkv.{bias, weight}
blocks.12.attn.proj.{bias, weight}
blocks.12.norm2.{bias, weight}
blocks.12.mlp.fc1.{bias, weight}
blocks.12.mlp.fc2.{bias, weight}
blocks.13.norm1.{bias, weight}
blocks.13.attn.qkv.{bias, weight}
blocks.13.attn.proj.{bias, weight}
blocks.13.norm2.{bias, weight}
blocks.13.mlp.fc1.{bias, weight}
blocks.13.mlp.fc2.{bias, weight}
blocks.14.norm1.{bias, weight}
blocks.14.attn.qkv.{bias, weight}
blocks.14.attn.proj.{bias, weight}
blocks.14.norm2.{bias, weight}
blocks.14.mlp.fc1.{bias, weight}
blocks.14.mlp.fc2.{bias, weight}
blocks.15.norm1.{bias, weight}
blocks.15.attn.qkv.{bias, weight}
blocks.15.attn.proj.{bias, weight}
blocks.15.norm2.{bias, weight}
blocks.15.mlp.fc1.{bias, weight}
blocks.15.mlp.fc2.{bias, weight}
blocks.16.norm1.{bias, weight}
blocks.16.attn.qkv.{bias, weight}
blocks.16.attn.proj.{bias, weight}
blocks.16.norm2.{bias, weight}
blocks.16.mlp.fc1.{bias, weight}
blocks.16.mlp.fc2.{bias, weight}
blocks.17.norm1.{bias, weight}
blocks.17.attn.qkv.{bias, weight}
blocks.17.attn.proj.{bias, weight}
blocks.17.norm2.{bias, weight}
blocks.17.mlp.fc1.{bias, weight}
blocks.17.mlp.fc2.{bias, weight}
blocks.18.norm1.{bias, weight}
blocks.18.attn.qkv.{bias, weight}
blocks.18.attn.proj.{bias, weight}
blocks.18.norm2.{bias, weight}
blocks.18.mlp.fc1.{bias, weight}
blocks.18.mlp.fc2.{bias, weight}
blocks.19.norm1.{bias, weight}
blocks.19.attn.qkv.{bias, weight}
blocks.19.attn.proj.{bias, weight}
blocks.19.norm2.{bias, weight}
blocks.19.mlp.fc1.{bias, weight}
blocks.19.mlp.fc2.{bias, weight}
blocks.20.norm1.{bias, weight}
blocks.20.attn.qkv.{bias, weight}
blocks.20.attn.proj.{bias, weight}
blocks.20.norm2.{bias, weight}
blocks.20.mlp.fc1.{bias, weight}
blocks.20.mlp.fc2.{bias, weight}
blocks.21.norm1.{bias, weight}
blocks.21.attn.qkv.{bias, weight}
blocks.21.attn.proj.{bias, weight}
blocks.21.norm2.{bias, weight}
blocks.21.mlp.fc1.{bias, weight}
blocks.21.mlp.fc2.{bias, weight}
blocks.22.norm1.{bias, weight}
blocks.22.attn.qkv.{bias, weight}
blocks.22.attn.proj.{bias, weight}
blocks.22.norm2.{bias, weight}
blocks.22.mlp.fc1.{bias, weight}
blocks.22.mlp.fc2.{bias, weight}
blocks.23.norm1.{bias, weight}
blocks.23.attn.qkv.{bias, weight}
blocks.23.attn.proj.{bias, weight}
blocks.23.norm2.{bias, weight}
blocks.23.mlp.fc1.{bias, weight}
blocks.23.mlp.fc2.{bias, weight}
norm.{bias, weight}

[ffxxy]

权重似乎不匹配

[ffxxy]

通过修改参数前缀后，可以成功加载，不过transformer部分的权重仍然随机初始化，导致模型的loss无法下降，请问如何获取transformer部分的权重呢。

还有一个问题，您更新后的mr_detr_pp_align，label的标签是从1开始还是0呢。

[ffxxy]

修改后的代码时基于DINO的，仓库中的权重无法使用了

[zhangchbin]

我来检查下

[ffxxy]

我来检查下

dino_swin_large_384_4scale_12ep.py的模型能够顺利训练了，显存溢出的问题也解决了，非常感谢您的贡献，但是vit版本的模型由于没有预训练权重，模型很难训练，3个epoch过后map仍然为0。

[zhangchbin]

我检查了一下huggingface上传的pretrained weights没什么问题。

可以使用 train.init_checkpoint=<file-path>来加载权重。

ViT的pretrained weights 可以使用EVA的预训练权重 https://github.com/IDEA-Research/detrex/tree/main/projects/dino_eva

[zhangchbin]

通过修改参数前缀后，可以成功加载，不过transformer部分的权重仍然随机初始化，导致模型的loss无法下降，请问如何获取transformer部分的权重呢。

还有一个问题，您更新后的mr_detr_pp_align，label的标签是从1开始还是0呢。

这个和之前的是一样的。

[ffxxy]

想再请教一下，为什么更新后的代码没有eva_02，是eva_02只支持输入为正方形吗

[zhangchbin]

不是正方形会有什么问题吗？

[ffxxy]

之前修改代码时，不是方形输入，这个模块前向传播会报错。

[ffxxy]

路径：detrex/modeling/backbone/eva_02.py

[zhangchbin]

可以padding成方形，直接在config里设置一样的宽和高就行，它也是用padding来实现的。

[ffxxy]

请问是把这里的

替换成T.ResizeScale吗

[zhangchbin]

参考这个就可以：
https://github.com/IDEA-Research/detrex/blob/c56d32e3d0262cff9835ebe80a0642965ae0cb3e/projects/dino_eva/configs/common/coco_loader_lsj.py#L8

[ffxxy]

请问这个权重可以直接加载的吗，需要修改模型结构吗

…

---原始邮件--- 发件人: "Chang-Bin ***@***.***&gt; 发送时间: 2025年8月8日(周五) 下午4:10 收件人: ***@***.***&gt;; 抄送: ***@***.******@***.***&gt;; 主题: Re: [Visual-AI/Mr.DETR] Mr.DETR++预训练权重加载问题 (Issue #16) zhangchbin left a comment (Visual-AI/Mr.DETR#16) Screenshot.2025-08-08.at.16.06.41.png (view on web) 我检查了一下huggingface上传的pretrained weights没什么问题。 可以使用 train.init_checkpoint=<file-path&gt;来加载权重。 ViT的pretrained weights 可以使用EVA的预训练权重 https://github.com/IDEA-Research/detrex/tree/main/projects/dino_eva — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: ***@***.***&gt;