refr-dfine-cuda

This model is a fine-tuned version of ustc-community/dfine-small-coco on the Mrtnl/refr-defect-detection dataset. It achieves the following results on the evaluation set:

• Loss: 6.7285
• Map: 0.026
• Map 50: 0.0682
• Map 75: 0.0144
• Map Small: -1.0
• Map Medium: 0.0267
• Map Large: 0.143
• Mar 1: 0.0333
• Mar 10: 0.25
• Mar 100: 0.5667
• Mar Small: -1.0
• Mar Medium: 0.525
• Mar Large: 0.65
• Map Defect: 0.026
• Mar 100 Defect: 0.5667

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• 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: 50.0

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 Defect	Mar 100 Defect
No log	1.0	3	61.5103	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	2.0	6	54.5423	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	3.0	9	46.0754	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	4.0	12	35.7740	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	5.0	15	29.1773	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	6.0	18	25.3934	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	7.0	21	23.8015	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	8.0	24	21.8589	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	9.0	27	19.4984	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	10.0	30	17.8712	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	11.0	33	16.2532	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	12.0	36	15.6115	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	13.0	39	14.6865	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0	0.0	-1.0	0.0	0.0	0.0	0.0
No log	14.0	42	14.1021	0.0001	0.0003	0.0	-1.0	0.0003	0.0	0.0	0.0	0.0333	-1.0	0.05	0.0	0.0001	0.0333
No log	15.0	45	13.4760	0.0003	0.0004	0.0004	-1.0	0.0013	0.0	0.0	0.0	0.1	-1.0	0.15	0.0	0.0003	0.1
No log	16.0	48	12.9603	0.0003	0.0005	0.0005	-1.0	0.0016	0.0	0.0	0.0	0.1	-1.0	0.15	0.0	0.0003	0.1
No log	17.0	51	12.1859	0.0002	0.0004	0.0004	-1.0	0.0016	0.0	0.0	0.0	0.1	-1.0	0.15	0.0	0.0002	0.1
No log	18.0	54	10.8084	0.0002	0.0004	0.0004	-1.0	0.0024	0.0	0.0	0.0	0.1	-1.0	0.15	0.0	0.0002	0.1
No log	19.0	57	10.2827	0.0002	0.0004	0.0004	-1.0	0.0017	0.0	0.0	0.0	0.1	-1.0	0.15	0.0	0.0002	0.1
No log	20.0	60	9.9501	0.0004	0.0013	0.0004	-1.0	0.0022	0.0	0.0	0.0	0.1333	-1.0	0.2	0.0	0.0004	0.1333
No log	21.0	63	9.8148	0.0009	0.0048	0.0003	-1.0	0.0021	0.0016	0.0	0.0	0.1833	-1.0	0.2	0.15	0.0009	0.1833
No log	22.0	66	9.5424	0.0021	0.0072	0.0012	-1.0	0.0037	0.0024	0.0	0.0	0.3167	-1.0	0.35	0.25	0.0021	0.3167
No log	23.0	69	9.1601	0.0028	0.0075	0.0015	-1.0	0.0056	0.003	0.0	0.0	0.3833	-1.0	0.425	0.3	0.0028	0.3833
No log	24.0	72	8.8512	0.0031	0.0089	0.0018	-1.0	0.0058	0.0031	0.0	0.0	0.3667	-1.0	0.4	0.3	0.0031	0.3667
No log	25.0	75	8.3892	0.0044	0.0151	0.0018	-1.0	0.0065	0.0089	0.0	0.0	0.4167	-1.0	0.45	0.35	0.0044	0.4167
No log	26.0	78	8.0770	0.0058	0.0162	0.0036	-1.0	0.0093	0.0104	0.0	0.0	0.45	-1.0	0.45	0.45	0.0058	0.45
No log	27.0	81	8.1041	0.0084	0.0259	0.0047	-1.0	0.0133	0.0152	0.0	0.0667	0.5167	-1.0	0.45	0.65	0.0084	0.5167
No log	28.0	84	8.0993	0.0083	0.0259	0.0029	-1.0	0.0136	0.0099	0.0	0.1	0.4333	-1.0	0.475	0.35	0.0083	0.4333
No log	29.0	87	7.9410	0.0111	0.0344	0.0021	-1.0	0.0191	0.0123	0.0	0.1167	0.4833	-1.0	0.55	0.35	0.0111	0.4833
No log	30.0	90	7.8057	0.0136	0.0351	0.0028	-1.0	0.0134	0.106	0.0	0.0833	0.5833	-1.0	0.575	0.6	0.0136	0.5833
No log	31.0	93	7.5607	0.0241	0.0617	0.0035	-1.0	0.0138	0.2083	0.0667	0.0833	0.5333	-1.0	0.475	0.65	0.0241	0.5333
No log	32.0	96	7.4293	0.0196	0.0503	0.0037	-1.0	0.0228	0.1066	0.0667	0.1333	0.65	-1.0	0.65	0.65	0.0196	0.65
No log	33.0	99	7.1287	0.0197	0.0455	0.0037	-1.0	0.0184	0.2083	0.0	0.1333	0.65	-1.0	0.65	0.65	0.0197	0.65
No log	34.0	102	7.0185	0.0204	0.0443	0.0057	-1.0	0.0194	0.2141	0.0	0.1333	0.6333	-1.0	0.625	0.65	0.0204	0.6333
No log	35.0	105	7.0300	0.0235	0.0569	0.007	-1.0	0.03	0.0754	0.0	0.3167	0.6167	-1.0	0.625	0.6	0.0235	0.6167
No log	36.0	108	7.0289	0.0166	0.0467	0.0061	-1.0	0.0213	0.0718	0.0	0.1167	0.5333	-1.0	0.5	0.6	0.0166	0.5333
No log	37.0	111	7.0824	0.0126	0.0394	0.0049	-1.0	0.0198	0.0472	0.0	0.1167	0.5	-1.0	0.45	0.6	0.0126	0.5
No log	38.0	114	7.0406	0.0136	0.0358	0.0049	-1.0	0.0209	0.0229	0.0	0.15	0.5167	-1.0	0.475	0.6	0.0136	0.5167
No log	39.0	117	7.0072	0.0152	0.0344	0.0075	-1.0	0.0206	0.0555	0.0	0.1333	0.5167	-1.0	0.475	0.6	0.0152	0.5167
No log	40.0	120	7.0165	0.0164	0.0449	0.0063	-1.0	0.0188	0.0412	0.0	0.1333	0.5167	-1.0	0.475	0.6	0.0164	0.5167
No log	41.0	123	6.9714	0.0199	0.0502	0.0065	-1.0	0.0225	0.0585	0.0	0.1833	0.55	-1.0	0.525	0.6	0.0199	0.55
No log	42.0	126	6.9177	0.0195	0.0505	0.0119	-1.0	0.0223	0.0793	0.0	0.1833	0.5667	-1.0	0.55	0.6	0.0195	0.5667
No log	43.0	129	6.9341	0.0186	0.0506	0.0056	-1.0	0.0196	0.0414	0.0	0.1833	0.5833	-1.0	0.575	0.6	0.0186	0.5833
No log	44.0	132	6.8606	0.0214	0.0523	0.0061	-1.0	0.0237	0.129	0.0333	0.3333	0.6167	-1.0	0.6	0.65	0.0214	0.6167
No log	45.0	135	6.7879	0.0236	0.0626	0.0133	-1.0	0.0251	0.1441	0.0	0.2667	0.6	-1.0	0.6	0.6	0.0236	0.6
No log	46.0	138	6.8562	0.0167	0.0453	0.004	-1.0	0.0165	0.0602	0.0	0.1167	0.55	-1.0	0.525	0.6	0.0167	0.55
No log	47.0	141	6.8504	0.0177	0.0499	0.0038	-1.0	0.0151	0.1296	0.0	0.1167	0.5333	-1.0	0.5	0.6	0.0177	0.5333
No log	48.0	144	6.8185	0.0195	0.0499	0.0111	-1.0	0.0192	0.1463	0.0	0.15	0.55	-1.0	0.525	0.6	0.0195	0.55
No log	49.0	147	6.7303	0.0263	0.0682	0.0144	-1.0	0.0269	0.1445	0.0333	0.25	0.5667	-1.0	0.525	0.65	0.0263	0.5667
No log	50.0	150	6.7514	0.023	0.0601	0.0128	-1.0	0.0237	0.1394	0.0333	0.1333	0.5667	-1.0	0.525	0.65	0.023	0.5667

Framework versions

• Transformers 5.10.1
• Pytorch 2.12.0+cu130
• Datasets 4.8.5
• Tokenizers 0.22.2