detr-resnet-50_fine_tuned_loc-2023

For new projects, consider small-models-for-glam/historic-newspaper-illustrations-yolov11 — a more recent YOLO-based detector for illustrations in historic newspaper pages, faster at inference. This DETR model is kept for reproducibility.

This model is a fine-tuned version of facebook/detr-resnet-50 on the loc_beyond_words dataset. It achieves the following results on the evaluation set:

Loss: 0.8784

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: 0.0001
train_batch_size: 8
eval_batch_size: 8
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss
3.731	0.16	50	2.6356
2.4875	0.31	100	2.2348
2.1786	0.47	150	2.1148
1.9845	0.62	200	1.8847
1.8507	0.78	250	1.8331
1.6813	0.94	300	1.5620
1.5613	1.09	350	1.5898
1.4966	1.25	400	1.4161
1.4831	1.41	450	1.4831
1.4587	1.56	500	1.3218
1.433	1.72	550	1.3529
1.33	1.88	600	1.2453
1.2842	2.03	650	1.2956
1.2807	2.19	700	1.1993
1.1767	2.34	750	1.1557
1.2134	2.5	800	1.1393
1.1897	2.66	850	1.2016
1.1784	2.81	900	1.1235
1.2016	2.97	950	1.1378
1.06	3.12	1000	1.0803
1.1124	3.28	1050	1.1145
1.1191	3.44	1100	1.0523
1.0819	3.59	1150	1.0165
1.1196	3.75	1200	1.0349
1.0534	3.91	1250	1.0441
1.0365	4.06	1300	1.1177
0.9853	4.22	1350	1.0721
0.9984	4.38	1400	0.9923
0.9802	4.53	1450	1.0079
1.04	4.69	1500	1.0198
1.098	4.84	1550	0.9788
1.079	5.0	1600	1.0291
1.0664	5.16	1650	0.9691
0.9715	5.31	1700	0.9380
0.9723	5.47	1750	1.0164
1.0019	5.62	1800	1.0064
0.9895	5.78	1850	1.0364
0.9835	5.94	1900	0.9848
0.994	6.09	1950	0.9353
0.9693	6.25	2000	0.9425
0.9413	6.41	2050	0.9173
0.9375	6.56	2100	0.9663
0.952	6.72	2150	0.8951
0.8927	6.88	2200	0.9099
0.8777	7.03	2250	0.9238
0.8976	7.19	2300	0.9715
0.9451	7.34	2350	0.9373
0.8972	7.5	2400	0.8959
0.9393	7.66	2450	1.0062
0.9	7.81	2500	0.8920
0.915	7.97	2550	0.8833
0.9018	8.12	2600	0.8671
0.8272	8.28	2650	0.9304
0.943	8.44	2700	0.8593
0.8667	8.59	2750	0.8875
0.871	8.75	2800	0.8457
0.9023	8.91	2850	0.8448
0.8733	9.06	2900	0.8261
0.8686	9.22	2950	0.8489
0.8412	9.38	3000	0.8244
0.8385	9.53	3050	0.8830
0.891	9.69	3100	0.8349
0.8692	9.84	3150	0.8672
0.8247	10.0	3200	0.8811
0.799	10.16	3250	0.8784

Framework versions

Transformers 4.27.4
Pytorch 2.0.0+cu118
Datasets 2.11.0
Tokenizers 0.13.3

Loading…

detr-resnet-50_fine_tuned_loc-2023

For new projects, consider small-models-for-glam/historic-newspaper-illustrations-yolov11 — a more recent YOLO-based detector for illustrations in historic newspaper pages, faster at inference. This DETR model is kept for reproducibility.

This model is a fine-tuned version of facebook/detr-resnet-50 on the loc_beyond_words dataset. It achieves the following results on the evaluation set:

Loss: 0.8784

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0001

train_batch_size: 8

eval_batch_size: 8

seed: 42

optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08

lr_scheduler_type: linear

num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss
3.731	0.16	50	2.6356
2.4875	0.31	100	2.2348
2.1786	0.47	150	2.1148
1.9845	0.62	200	1.8847
1.8507	0.78	250	1.8331
1.6813	0.94	300	1.5620
1.5613	1.09	350	1.5898
1.4966	1.25	400	1.4161
1.4831	1.41	450	1.4831
1.4587	1.56	500	1.3218
1.433	1.72	550	1.3529
1.33	1.88	600	1.2453
1.2842	2.03	650	1.2956
1.2807	2.19	700	1.1993
1.1767	2.34	750	1.1557
1.2134	2.5	800	1.1393
1.1897	2.66	850	1.2016
1.1784	2.81	900	1.1235
1.2016	2.97	950	1.1378
1.06	3.12	1000	1.0803
1.1124	3.28	1050	1.1145
1.1191	3.44	1100	1.0523
1.0819	3.59	1150	1.0165
1.1196	3.75	1200	1.0349
1.0534	3.91	1250	1.0441
1.0365	4.06	1300	1.1177
0.9853	4.22	1350	1.0721
0.9984	4.38	1400	0.9923
0.9802	4.53	1450	1.0079
1.04	4.69	1500	1.0198
1.098	4.84	1550	0.9788
1.079	5.0	1600	1.0291
1.0664	5.16	1650	0.9691
0.9715	5.31	1700	0.9380
0.9723	5.47	1750	1.0164
1.0019	5.62	1800	1.0064
0.9895	5.78	1850	1.0364
0.9835	5.94	1900	0.9848
0.994	6.09	1950	0.9353
0.9693	6.25	2000	0.9425
0.9413	6.41	2050	0.9173
0.9375	6.56	2100	0.9663
0.952	6.72	2150	0.8951
0.8927	6.88	2200	0.9099
0.8777	7.03	2250	0.9238
0.8976	7.19	2300	0.9715
0.9451	7.34	2350	0.9373
0.8972	7.5	2400	0.8959
0.9393	7.66	2450	1.0062
0.9	7.81	2500	0.8920
0.915	7.97	2550	0.8833
0.9018	8.12	2600	0.8671
0.8272	8.28	2650	0.9304
0.943	8.44	2700	0.8593
0.8667	8.59	2750	0.8875
0.871	8.75	2800	0.8457
0.9023	8.91	2850	0.8448
0.8733	9.06	2900	0.8261
0.8686	9.22	2950	0.8489
0.8412	9.38	3000	0.8244
0.8385	9.53	3050	0.8830
0.891	9.69	3100	0.8349
0.8692	9.84	3150	0.8672
0.8247	10.0	3200	0.8811
0.799	10.16	3250	0.8784

Framework versions

Transformers 4.27.4

Pytorch 2.0.0+cu118

Datasets 2.11.0

Tokenizers 0.13.3

biglam/detr-resnet-50_fine_tuned_loc-2023

detr-resnet-50_fine_tuned_loc-2023

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

detr-resnet-50_fine_tuned_loc-2023

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions