segformer-b0-finetuned-segments-sidewalk-oct-22

This model is a fine-tuned version of nvidia/mit-b0 on the segments/sidewalk-semantic dataset. It achieves the following results on the evaluation set:

• Loss: 0.5041
• Mean Iou: 0.3425
• Mean Accuracy: 0.4000
• Overall Accuracy: 0.8735
• Accuracy Unlabeled: nan
• Accuracy Flat-road: 0.8915
• Accuracy Flat-sidewalk: 0.9632
• Accuracy Flat-crosswalk: 0.5369
• Accuracy Flat-cyclinglane: 0.8945
• Accuracy Flat-parkingdriveway: 0.5757
• Accuracy Flat-railtrack: nan
• Accuracy Flat-curb: 0.6197
• Accuracy Human-person: 0.8098
• Accuracy Human-rider: 0.0
• Accuracy Vehicle-car: 0.9559
• Accuracy Vehicle-truck: 0.0
• Accuracy Vehicle-bus: 0.0
• Accuracy Vehicle-tramtrain: nan
• Accuracy Vehicle-motorcycle: 0.0
• Accuracy Vehicle-bicycle: 0.5943
• Accuracy Vehicle-caravan: 0.0
• Accuracy Vehicle-cartrailer: 0.0
• Accuracy Construction-building: 0.8870
• Accuracy Construction-door: 0.0755
• Accuracy Construction-wall: 0.5240
• Accuracy Construction-fenceguardrail: 0.4137
• Accuracy Construction-bridge: 0.0
• Accuracy Construction-tunnel: nan
• Accuracy Construction-stairs: 0.0
• Accuracy Object-pole: 0.4666
• Accuracy Object-trafficsign: 0.0
• Accuracy Object-trafficlight: 0.0
• Accuracy Nature-vegetation: 0.9278
• Accuracy Nature-terrain: 0.8655
• Accuracy Sky: 0.9629
• Accuracy Void-ground: 0.0197
• Accuracy Void-dynamic: 0.0120
• Accuracy Void-static: 0.4024
• Accuracy Void-unclear: 0.0
• Iou Unlabeled: nan
• Iou Flat-road: 0.7833
• Iou Flat-sidewalk: 0.8989
• Iou Flat-crosswalk: 0.4792
• Iou Flat-cyclinglane: 0.8079
• Iou Flat-parkingdriveway: 0.4573
• Iou Flat-railtrack: nan
• Iou Flat-curb: 0.4863
• Iou Human-person: 0.6326
• Iou Human-rider: 0.0
• Iou Vehicle-car: 0.8576
• Iou Vehicle-truck: 0.0
• Iou Vehicle-bus: 0.0
• Iou Vehicle-tramtrain: nan
• Iou Vehicle-motorcycle: 0.0
• Iou Vehicle-bicycle: 0.5047
• Iou Vehicle-caravan: 0.0
• Iou Vehicle-cartrailer: 0.0
• Iou Construction-building: 0.7497
• Iou Construction-door: 0.0679
• Iou Construction-wall: 0.3758
• Iou Construction-fenceguardrail: 0.3306
• Iou Construction-bridge: 0.0
• Iou Construction-tunnel: nan
• Iou Construction-stairs: 0.0
• Iou Object-pole: 0.3653
• Iou Object-trafficsign: 0.0
• Iou Object-trafficlight: 0.0
• Iou Nature-vegetation: 0.8475
• Iou Nature-terrain: 0.7425
• Iou Sky: 0.9155
• Iou Void-ground: 0.0144
• Iou Void-dynamic: 0.0056
• Iou Void-static: 0.2942
• Iou Void-unclear: 0.0

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Unlabeled	Accuracy Flat-road	Accuracy Flat-sidewalk	Accuracy Flat-crosswalk	Accuracy Flat-cyclinglane	Accuracy Flat-parkingdriveway	Accuracy Flat-railtrack	Accuracy Flat-curb	Accuracy Human-person	Accuracy Human-rider	Accuracy Vehicle-car	Accuracy Vehicle-truck	Accuracy Vehicle-bus	Accuracy Vehicle-tramtrain	Accuracy Vehicle-motorcycle	Accuracy Vehicle-bicycle	Accuracy Vehicle-caravan	Accuracy Vehicle-cartrailer	Accuracy Construction-building	Accuracy Construction-door	Accuracy Construction-wall	Accuracy Construction-fenceguardrail	Accuracy Construction-bridge	Accuracy Construction-tunnel	Accuracy Construction-stairs	Accuracy Object-pole	Accuracy Object-trafficsign	Accuracy Object-trafficlight	Accuracy Nature-vegetation	Accuracy Nature-terrain	Accuracy Sky	Accuracy Void-ground	Accuracy Void-dynamic	Accuracy Void-static	Accuracy Void-unclear	Iou Unlabeled	Iou Flat-road	Iou Flat-sidewalk	Iou Flat-crosswalk	Iou Flat-cyclinglane	Iou Flat-parkingdriveway	Iou Flat-railtrack	Iou Flat-curb	Iou Human-person	Iou Human-rider	Iou Vehicle-car	Iou Vehicle-truck	Iou Vehicle-bus	Iou Vehicle-tramtrain	Iou Vehicle-motorcycle	Iou Vehicle-bicycle	Iou Vehicle-caravan	Iou Vehicle-cartrailer	Iou Construction-building	Iou Construction-door	Iou Construction-wall	Iou Construction-fenceguardrail	Iou Construction-bridge	Iou Construction-tunnel	Iou Construction-stairs	Iou Object-pole	Iou Object-trafficsign	Iou Object-trafficlight	Iou Nature-vegetation	Iou Nature-terrain	Iou Sky	Iou Void-ground	Iou Void-dynamic	Iou Void-static	Iou Void-unclear
0.8295	10.0	500	0.5784	0.2968	0.3454	0.8488	nan	0.8644	0.9708	0.5489	0.8284	0.3579	nan	0.4601	0.5972	0.0	0.9478	0.0	0.0	nan	0.0	0.2258	0.0	0.0	0.8755	0.0	0.5088	0.2398	0.0	nan	0.0	0.2850	0.0	0.0	0.9202	0.8444	0.9562	0.0	0.0	0.2760	0.0	nan	0.7422	0.8610	0.5062	0.7465	0.3036	nan	0.3674	0.4975	0.0	0.7854	0.0	0.0	nan	0.0	0.2219	0.0	0.0	0.7154	0.0	0.3535	0.2111	0.0	nan	0.0	0.2336	0.0	0.0	0.8300	0.7174	0.8829	0.0	0.0	0.2252	0.0
0.3353	20.0	1000	0.4984	0.3285	0.3821	0.8672	nan	0.8909	0.9622	0.6104	0.8725	0.5427	nan	0.5603	0.8060	0.0	0.9490	0.0	0.0	nan	0.0	0.5338	0.0	0.0	0.9165	0.0	0.4398	0.2810	0.0	nan	0.0	0.3870	0.0	0.0	0.9239	0.8398	0.9537	0.0	0.0	0.3751	0.0	nan	0.7803	0.8865	0.5464	0.8127	0.4178	nan	0.4462	0.6088	0.0	0.8397	0.0	0.0	nan	0.0	0.4671	0.0	0.0	0.7292	0.0	0.3445	0.2352	0.0	nan	0.0	0.3071	0.0	0.0	0.8419	0.7374	0.9037	0.0	0.0	0.2799	0.0
0.2724	30.0	1500	0.5030	0.3375	0.3970	0.8699	nan	0.8602	0.9653	0.5857	0.8991	0.5570	nan	0.6176	0.8335	0.0	0.9490	0.0	0.0	nan	0.0	0.6118	0.0	0.0	0.8834	0.0	0.5454	0.4152	0.0	nan	0.0	0.4468	0.0	0.0	0.9323	0.8584	0.9619	0.0129	0.0044	0.3683	0.0	nan	0.7672	0.8932	0.5331	0.7681	0.4659	nan	0.4744	0.6194	0.0	0.8559	0.0	0.0	nan	0.0	0.5150	0.0	0.0	0.7431	0.0	0.3743	0.3207	0.0	nan	0.0	0.3420	0.0	0.0	0.8460	0.7431	0.9113	0.0109	0.0024	0.2756	0.0
0.2573	40.0	2000	0.5071	0.3410	0.4013	0.8711	nan	0.8665	0.9641	0.5698	0.9008	0.5656	nan	0.6239	0.8238	0.0	0.9554	0.0	0.0	nan	0.0	0.6063	0.0	0.0	0.8794	0.0491	0.5325	0.4088	0.0	nan	0.0	0.4805	0.0	0.0	0.9293	0.8622	0.9639	0.0134	0.0089	0.4358	0.0	nan	0.7735	0.8969	0.5057	0.7822	0.4407	nan	0.4920	0.6306	0.0	0.8590	0.0	0.0	nan	0.0	0.4998	0.0	0.0	0.7457	0.0451	0.3710	0.3274	0.0	nan	0.0	0.3690	0.0	0.0	0.8468	0.7403	0.9150	0.0099	0.0049	0.3153	0.0
0.2997	50.0	2500	0.5041	0.3425	0.4000	0.8735	nan	0.8915	0.9632	0.5369	0.8945	0.5757	nan	0.6197	0.8098	0.0	0.9559	0.0	0.0	nan	0.0	0.5943	0.0	0.0	0.8870	0.0755	0.5240	0.4137	0.0	nan	0.0	0.4666	0.0	0.0	0.9278	0.8655	0.9629	0.0197	0.0120	0.4024	0.0	nan	0.7833	0.8989	0.4792	0.8079	0.4573	nan	0.4863	0.6326	0.0	0.8576	0.0	0.0	nan	0.0	0.5047	0.0	0.0	0.7497	0.0679	0.3758	0.3306	0.0	nan	0.0	0.3653	0.0	0.0	0.8475	0.7425	0.9155	0.0144	0.0056	0.2942	0.0

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1+cu118
• Datasets 2.17.1
• Tokenizers 0.15.2