Edit model card

ecc_segformerv2

This model is a fine-tuned version of nvidia/mit-b5 on the rishitunu/ecc_crackdetector_dataset dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3478
  • Mean Iou: 0.0862
  • Mean Accuracy: 0.1924
  • Overall Accuracy: 0.1924
  • Accuracy Background: nan
  • Accuracy Crack: 0.1924
  • Iou Background: 0.0
  • Iou Crack: 0.1723

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: 6e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 1337
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: polynomial
  • training_steps: 10000

Training results

Training Loss Epoch Step Validation Loss Mean Iou Mean Accuracy Overall Accuracy Accuracy Background Accuracy Crack Iou Background Iou Crack
0.1019 1.0 251 0.5116 0.1490 0.3280 0.3280 nan 0.3280 0.0 0.2979
0.0938 2.0 502 0.4725 0.1144 0.2400 0.2400 nan 0.2400 0.0 0.2287
0.098 3.0 753 0.5117 0.1276 0.2748 0.2748 nan 0.2748 0.0 0.2552
0.1018 4.0 1004 0.3870 0.1053 0.2254 0.2254 nan 0.2254 0.0 0.2106
0.0928 5.0 1255 0.2907 0.0772 0.1630 0.1630 nan 0.1630 0.0 0.1544
0.0936 6.0 1506 0.5220 0.1193 0.2544 0.2544 nan 0.2544 0.0 0.2385
0.077 7.0 1757 0.1608 0.0617 0.1308 0.1308 nan 0.1308 0.0 0.1235
0.0963 8.0 2008 0.1756 0.0456 0.0923 0.0923 nan 0.0923 0.0 0.0912
0.0958 9.0 2259 0.2027 0.0862 0.1813 0.1813 nan 0.1813 0.0 0.1725
0.0755 10.0 2510 0.2327 0.0888 0.1832 0.1832 nan 0.1832 0.0 0.1776
0.0632 11.0 2761 0.2169 0.0846 0.1863 0.1863 nan 0.1863 0.0 0.1693
0.0638 12.0 3012 0.2309 0.0852 0.1957 0.1957 nan 0.1957 0.0 0.1704
0.0509 13.0 3263 0.3209 0.1236 0.2910 0.2910 nan 0.2910 0.0 0.2472
0.0497 14.0 3514 0.3274 0.1045 0.2354 0.2354 nan 0.2354 0.0 0.2089
0.0396 15.0 3765 0.3415 0.1005 0.2257 0.2257 nan 0.2257 0.0 0.2010
0.0373 16.0 4016 0.3530 0.1122 0.2486 0.2486 nan 0.2486 0.0 0.2244
0.0388 17.0 4267 0.3312 0.0889 0.1974 0.1974 nan 0.1974 0.0 0.1778
0.0346 18.0 4518 0.3061 0.0903 0.2125 0.2125 nan 0.2125 0.0 0.1807
0.0296 19.0 4769 0.3223 0.1000 0.2315 0.2315 nan 0.2315 0.0 0.2000
0.0311 20.0 5020 0.3458 0.0943 0.2237 0.2237 nan 0.2237 0.0 0.1887
0.0303 21.0 5271 0.3283 0.0975 0.2255 0.2255 nan 0.2255 0.0 0.1951
0.0249 22.0 5522 0.3387 0.0998 0.2327 0.2327 nan 0.2327 0.0 0.1996
0.0298 23.0 5773 0.3332 0.0973 0.2242 0.2242 nan 0.2242 0.0 0.1946
0.0239 24.0 6024 0.3778 0.1146 0.2634 0.2634 nan 0.2634 0.0 0.2292
0.0238 25.0 6275 0.3250 0.0909 0.2081 0.2081 nan 0.2081 0.0 0.1818
0.0242 26.0 6526 0.3826 0.1002 0.2285 0.2285 nan 0.2285 0.0 0.2004
0.017 27.0 6777 0.3543 0.1058 0.2367 0.2367 nan 0.2367 0.0 0.2115
0.0241 28.0 7028 0.3491 0.0915 0.2069 0.2069 nan 0.2069 0.0 0.1830
0.0203 29.0 7279 0.3354 0.0899 0.2056 0.2056 nan 0.2056 0.0 0.1798
0.0206 30.0 7530 0.3592 0.0944 0.2165 0.2165 nan 0.2165 0.0 0.1888
0.0211 31.0 7781 0.3200 0.0943 0.2100 0.2100 nan 0.2100 0.0 0.1886
0.0209 32.0 8032 0.3401 0.0850 0.1941 0.1941 nan 0.1941 0.0 0.1701
0.0172 33.0 8283 0.3326 0.0879 0.1986 0.1986 nan 0.1986 0.0 0.1759
0.0187 34.0 8534 0.3343 0.0869 0.1960 0.1960 nan 0.1960 0.0 0.1739
0.0181 35.0 8785 0.3223 0.0824 0.1835 0.1835 nan 0.1835 0.0 0.1648
0.0168 36.0 9036 0.3461 0.0864 0.1933 0.1933 nan 0.1933 0.0 0.1727
0.0169 37.0 9287 0.3438 0.0848 0.1888 0.1888 nan 0.1888 0.0 0.1695
0.0182 38.0 9538 0.3506 0.0865 0.1933 0.1933 nan 0.1933 0.0 0.1730
0.0167 39.0 9789 0.3535 0.0869 0.1946 0.1946 nan 0.1946 0.0 0.1739
0.0174 39.84 10000 0.3478 0.0862 0.1924 0.1924 nan 0.1924 0.0 0.1723

Framework versions

  • Transformers 4.32.0.dev0
  • Pytorch 2.0.1+cpu
  • Datasets 2.14.4
  • Tokenizers 0.13.3
Downloads last month
6
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for rishitunu/ecc_segformerv2

Base model

nvidia/mit-b5
Finetuned
(42)
this model