Some model parameters are not in the checkpoint?

[LIYHUI]

I use model_syn_r101_pretrain.pth as the pretrained model, and finetune on the Icdar2015 dataset. But I find that some model parameters are not in the checkpoint.
[08/07 10:01:59] fvcore.common.checkpoint INFO: Loading checkpoint from ./models/TextFuseNet_r101_model/model_syn_r101_pretrain.pth [08/07 10:02:01] fvcore.common.checkpoint INFO: Some model parameters are not in the checkpoint: �[34mroi_heads.seg_head.conv1x1_list.0.weight�[0m �[34mroi_heads.seg_head.conv1x1_list.1.weight�[0m �[34mroi_heads.seg_head.conv1x1_list.2.weight�[0m �[34mroi_heads.seg_head.conv1x1_list.3.weight�[0m �[34mroi_heads.seg_head.conv3x3_list.0.weight�[0m �[34mroi_heads.seg_head.conv3x3_list.1.weight�[0m �[34mroi_heads.seg_head.conv3x3_list.2.weight�[0m �[34mroi_heads.seg_head.conv3x3_list.3.weight�[0m �[34mroi_heads.seg_head.conv3x3_list_roi.0.weight�[0m �[34mroi_heads.seg_head.conv3x3_list_roi.1.weight�[0m �[34mroi_heads.seg_head.conv3x3_list_roi.2.weight�[0m �[34mroi_heads.seg_head.conv3x3_list_roi.3.weight�[0m �[34mroi_heads.seg_head.conv1x1_seg_logits.weight�[0m �[34mroi_heads.seg_head.seg_logits.{weight, bias}�[0m �[34mroi_heads.mutil_path_fuse_module.char_conv3x3.weight�[0m �[34mroi_heads.mutil_path_fuse_module.char_conv1x1.weight�[0m �[34mroi_heads.mutil_path_fuse_module.text_conv3x3.weight�[0m �[34mroi_heads.mutil_path_fuse_module.text_conv1x1.weight�[0m �[34mroi_heads.mutil_path_fuse_module.conv3x3.weight�[0m �[34mroi_heads.mutil_path_fuse_module.conv1x1.weight�[0m �[34mroi_heads.mutil_path_fuse_module.bn.{weight, bias, running_mean, running_var}�[0m [08/07 10:02:01] d2.engine.train_loop INFO: Starting training from iteration 0

why model_syn_r101_pretrain.pth differs from model_ic15_r101.pth?

[LIYHUI]

@ying09 @Real-YeJ Can you answer my confusion? Thanks~

[LIYHUI]

@ying09 请问可以解答一下我的疑惑吗？为什么SynthText预训练模型的结构和 model_ic15_r101.pth 是不一样的？

[ying09]

不好意思！这工作时间有点久远，做这个工作的同学也已经毕业，我会尽快让叶同学来回复你的问题。 liujuhua ***@***.***

…

---- Replied Message ***@***.***>Date08/14/2022 10:42 ***@***.***> CcJuhua ***@***.******@***.***>SubjectRe: [ying09/TextFuseNet] Some model parameters are not in the checkpoint? (Issue #100) @ying09 请问可以解答一下我的疑惑吗？为什么SynthText预训练模型的结构和 model_ic15_r101.pth 是不一样的？ —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***> [ { ***@***.***": "http://schema.org", ***@***.***": "EmailMessage", "potentialAction": { ***@***.***": "ViewAction", "target": "#100 (comment)", "url": "#100 (comment)", "name": "View Issue" }, "description": "View this Issue on GitHub", "publisher": { ***@***.***": "Organization", "name": "GitHub", "url": "https://github.com" } } ]

[LIYHUI]

好的，麻烦您了~