dropoff-utcustom-train-SF-RGB-b0_1

This model is a fine-tuned version of nvidia/mit-b0 on the sam1120/dropoff-utcustom-TRAIN dataset. It achieves the following results on the evaluation set:

• Loss: 0.5626
• Mean Iou: 0.4261
• Mean Accuracy: 0.7046
• Overall Accuracy: 0.9598
• Accuracy Unlabeled: nan
• Accuracy Dropoff: 0.4247
• Accuracy Undropoff: 0.9846
• Iou Unlabeled: 0.0
• Iou Dropoff: 0.3192
• Iou Undropoff: 0.9590

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: 9e-06
• 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
• lr_scheduler_warmup_ratio: 0.05
• num_epochs: 120

Training results

Training Loss	Epoch	Step	Validation Loss	Mean Iou	Mean Accuracy	Overall Accuracy	Accuracy Unlabeled	Accuracy Dropoff	Accuracy Undropoff	Iou Unlabeled	Iou Dropoff	Iou Undropoff
1.1029	3.33	10	1.0852	0.1637	0.3955	0.4522	nan	0.3333	0.4577	0.0	0.0410	0.4501
1.0856	6.67	20	1.0764	0.1911	0.5086	0.5025	nan	0.5153	0.5019	0.0	0.0761	0.4972
1.0755	10.0	30	1.0611	0.2252	0.6367	0.5749	nan	0.7045	0.5688	0.0	0.1104	0.5652
1.0285	13.33	40	1.0382	0.2622	0.7487	0.6568	nan	0.8494	0.6479	0.0	0.1420	0.6445
0.9935	16.67	50	1.0151	0.2893	0.7814	0.7201	nan	0.8486	0.7141	0.0	0.1580	0.7099
0.9927	20.0	60	0.9834	0.3160	0.7963	0.7816	nan	0.8124	0.7801	0.0	0.1735	0.7744
0.938	23.33	70	0.9585	0.3308	0.8084	0.8127	nan	0.8036	0.8131	0.0	0.1860	0.8065
0.9169	26.67	80	0.9376	0.3457	0.8169	0.8376	nan	0.7943	0.8396	0.0	0.2048	0.8324
0.8814	30.0	90	0.9003	0.3624	0.8086	0.8691	nan	0.7421	0.8750	0.0	0.2220	0.8651
0.8618	33.33	100	0.8894	0.3669	0.8184	0.8761	nan	0.7550	0.8817	0.0	0.2287	0.8720
0.8388	36.67	110	0.8618	0.3774	0.8096	0.8926	nan	0.7187	0.9006	0.0	0.2431	0.8892
0.8878	40.0	120	0.8269	0.3929	0.7937	0.9140	nan	0.6618	0.9257	0.0	0.2671	0.9116
0.8066	43.33	130	0.8074	0.4014	0.7955	0.9225	nan	0.6562	0.9348	0.0	0.2839	0.9202
0.8084	46.67	140	0.7919	0.4023	0.7932	0.9248	nan	0.6487	0.9376	0.0	0.2844	0.9226
0.7415	50.0	150	0.7707	0.4068	0.7850	0.9309	nan	0.6249	0.9451	0.0	0.2913	0.9290
0.7508	53.33	160	0.7326	0.4154	0.7660	0.9415	nan	0.5735	0.9585	0.0	0.3063	0.9400
0.7312	56.67	170	0.7126	0.4196	0.7636	0.9449	nan	0.5646	0.9625	0.0	0.3155	0.9435
0.6442	60.0	180	0.6869	0.4255	0.7500	0.9509	nan	0.5296	0.9704	0.0	0.3268	0.9497
0.6633	63.33	190	0.6765	0.4286	0.7524	0.9525	nan	0.5328	0.9719	0.0	0.3343	0.9513
0.7247	66.67	200	0.6557	0.4307	0.7335	0.9568	nan	0.4886	0.9785	0.0	0.3364	0.9558
0.6133	70.0	210	0.6369	0.4298	0.7279	0.9573	nan	0.4761	0.9796	0.0	0.3330	0.9564
0.6309	73.33	220	0.6309	0.4298	0.7437	0.9547	nan	0.5123	0.9752	0.0	0.3356	0.9536
0.6373	76.67	230	0.6094	0.4276	0.7197	0.9577	nan	0.4585	0.9808	0.0	0.3262	0.9568
0.8436	80.0	240	0.6195	0.4341	0.7438	0.9569	nan	0.5101	0.9776	0.0	0.3463	0.9559
0.6172	83.33	250	0.6207	0.4323	0.7384	0.9570	nan	0.4987	0.9782	0.0	0.3409	0.9560
0.6048	86.67	260	0.5949	0.4272	0.7136	0.9586	nan	0.4449	0.9824	0.0	0.3237	0.9578
0.7887	90.0	270	0.6007	0.4308	0.7282	0.9580	nan	0.4760	0.9803	0.0	0.3353	0.9571
0.605	93.33	280	0.5883	0.4284	0.7157	0.9589	nan	0.4489	0.9825	0.0	0.3271	0.9581
0.5964	96.67	290	0.5872	0.4277	0.7134	0.9590	nan	0.4439	0.9828	0.0	0.3251	0.9581
0.6097	100.0	300	0.5903	0.4300	0.7240	0.9582	nan	0.4669	0.9810	0.0	0.3325	0.9573
0.5886	103.33	310	0.5710	0.4250	0.7035	0.9594	nan	0.4227	0.9843	0.0	0.3162	0.9586
0.6079	106.67	320	0.5695	0.4277	0.7112	0.9594	nan	0.4390	0.9835	0.0	0.3245	0.9586
0.8054	110.0	330	0.5746	0.4308	0.7237	0.9588	nan	0.4657	0.9816	0.0	0.3344	0.9579
0.5496	113.33	340	0.5631	0.4285	0.7129	0.9595	nan	0.4424	0.9835	0.0	0.3269	0.9587
0.6271	116.67	350	0.5761	0.4302	0.7214	0.9589	nan	0.4608	0.9819	0.0	0.3326	0.9580
0.5511	120.0	360	0.5626	0.4261	0.7046	0.9598	nan	0.4247	0.9846	0.0	0.3192	0.9590

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3