manga-face-detector-dfine-v1

This model is a fine-tuned version of Attach3/manga-panel-detector-dfine on the Attach3/manga109s-faces-single dataset. It achieves the following results on the evaluation set:

• Loss: 4.5347
• Map: 0.2875
• Map 50: 0.5672
• Map 75: 0.2518
• Map Small: 0.1438
• Map Medium: 0.3099
• Map Large: 0.3967
• Mar 1: 0.0968
• Mar 10: 0.4484
• Mar 100: 0.5884
• Mar Small: 0.3936
• Mar Medium: 0.5848
• Mar Large: 0.7378
• Map Face: 0.2875
• Mar 100 Face: 0.5884

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 8
• eval_batch_size: 1
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 15.0

Training results

Training Loss	Epoch	Step	Validation Loss	Map	Map 50	Map 75	Map Small	Map Medium	Map Large	Mar 1	Mar 10	Mar 100	Mar Small	Mar Medium	Mar Large	Map Face	Mar 100 Face
15.4498	1.0	1383	3.7752	0.26	0.4968	0.2403	0.099	0.2599	0.3989	0.093	0.3632	0.5589	0.3394	0.5709	0.6845	0.26	0.5589
14.6906	2.0	2766	3.9415	0.2835	0.5478	0.2532	0.1221	0.2919	0.3987	0.0978	0.4086	0.5734	0.3696	0.5767	0.7106	0.2835	0.5734
14.4156	3.0	4149	4.0874	0.2964	0.5837	0.2582	0.133	0.3083	0.4032	0.0993	0.43	0.576	0.3759	0.5766	0.7179	0.2964	0.576
14.2519	4.0	5532	4.1849	0.3009	0.5961	0.2602	0.1386	0.3205	0.404	0.0997	0.4389	0.5777	0.3824	0.5766	0.7208	0.3009	0.5777
14.1930	5.0	6915	4.2457	0.2981	0.5859	0.2595	0.1408	0.3118	0.4042	0.0995	0.442	0.5811	0.3831	0.5795	0.7271	0.2981	0.5811
14.1194	6.0	8298	4.3367	0.2962	0.5865	0.2576	0.1394	0.3136	0.4029	0.0987	0.4457	0.5853	0.3932	0.5831	0.7288	0.2962	0.5853
14.0494	7.0	9681	4.3983	0.2958	0.5844	0.2598	0.1425	0.3173	0.4022	0.0979	0.4475	0.5852	0.3912	0.5822	0.7325	0.2958	0.5852
13.9709	8.0	11064	4.4487	0.2855	0.5689	0.2475	0.1401	0.3044	0.3898	0.0969	0.4444	0.5866	0.3911	0.5843	0.7327	0.2855	0.5866
13.9378	9.0	12447	4.4390	0.288	0.569	0.2536	0.1405	0.3065	0.3948	0.0977	0.4468	0.5861	0.3883	0.5837	0.7345	0.288	0.5861
13.9212	10.0	13830	4.4660	0.2911	0.5752	0.2571	0.1421	0.3113	0.4003	0.0979	0.4478	0.5865	0.3903	0.5832	0.7358	0.2911	0.5865
13.9169	11.0	15213	4.5232	0.2848	0.5656	0.2458	0.1408	0.3065	0.3864	0.0962	0.4448	0.585	0.3894	0.5806	0.7368	0.2848	0.585
13.8883	12.0	16596	4.5203	0.2846	0.5607	0.2505	0.1403	0.3058	0.3928	0.0971	0.4463	0.5891	0.3938	0.586	0.7373	0.2846	0.5891
13.9033	13.0	17979	4.5300	0.2934	0.5764	0.258	0.1422	0.3115	0.4047	0.0981	0.4507	0.5889	0.3951	0.5855	0.7367	0.2934	0.5889
13.8875	14.0	19362	4.5289	0.2863	0.5628	0.2509	0.1424	0.3076	0.3954	0.0968	0.4456	0.5872	0.3927	0.5833	0.7367	0.2863	0.5872
13.8552	15.0	20745	4.5347	0.2875	0.5672	0.2518	0.1438	0.3099	0.3967	0.0968	0.4484	0.5884	0.3936	0.5848	0.7378	0.2875	0.5884

Framework versions

• Transformers 5.13.0
• Pytorch 2.13.0+cu130
• Datasets 5.0.0
• Tokenizers 0.22.2