deepghs/eattach_monochrome_experiments

This is an experiment on DeepGHS's new tagger attachment training framework deepghs/emb_attachments.

Just some technical validation, not recommended for use in production.

(2025.3.7: This feature is still WIP, you can take a look at this PR and this documentation, you will like it.)

Name	Encoder	Problem	FLOPS	Params	AUC	accuracy	loss	mAP	Labels
mlp_layer1_seed114514_dist8	wdtagger:SmilingWolf/wd-swinv2-tagger-v3	classification	525.3K	525.8K	0.9937	0.9743	0.04483	0.9937	monochrome, normal
mlp_layer1_seedx_dist8	wdtagger:SmilingWolf/wd-swinv2-tagger-v3	classification	525.3K	525.8K	0.9946	0.9732	0.04058	0.9947	monochrome, normal
mlp_layer1_seed3_dist8	wdtagger:SmilingWolf/wd-swinv2-tagger-v3	classification	525.3K	525.8K	0.9956	0.9699	0.05301	0.9957	monochrome, normal
mlp_layer1_seed1	wdtagger:SmilingWolf/wd-swinv2-tagger-v3	classification	525.3K	525.8K	0.9948	0.9652	0.06904	0.995	monochrome, normal