Edit model card

swinv2-large-patch4-window12to16-192to256-22kto1k-ft-finetuned-galaxy10-decals

This model is a fine-tuned version of microsoft/swinv2-large-patch4-window12to16-192to256-22kto1k-ft on the matthieulel/galaxy10_decals dataset. It achieves the following results on the evaluation set:

  • Loss: 0.5226
  • Accuracy: 0.8591
  • Precision: 0.8571
  • Recall: 0.8591
  • F1: 0.8567

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: 64
  • eval_batch_size: 64
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 256
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_ratio: 0.1
  • num_epochs: 30

Training results

Training Loss Epoch Step Validation Loss Accuracy Precision Recall F1
1.0846 0.99 62 0.8092 0.7272 0.7246 0.7272 0.7101
0.7867 2.0 125 0.6366 0.7988 0.7996 0.7988 0.7895
0.6835 2.99 187 0.5315 0.8207 0.8195 0.8207 0.8157
0.586 4.0 250 0.4611 0.8489 0.8468 0.8489 0.8452
0.5263 4.99 312 0.4753 0.8399 0.8421 0.8399 0.8400
0.5341 6.0 375 0.4551 0.8427 0.8433 0.8427 0.8386
0.4743 6.99 437 0.4639 0.8382 0.8433 0.8382 0.8391
0.4573 8.0 500 0.4771 0.8360 0.8422 0.8360 0.8345
0.4368 8.99 562 0.4731 0.8472 0.8450 0.8472 0.8452
0.4022 10.0 625 0.4736 0.8540 0.8528 0.8540 0.8516
0.4005 10.99 687 0.4542 0.8551 0.8554 0.8551 0.8547
0.3514 12.0 750 0.5543 0.8467 0.8527 0.8467 0.8471
0.3565 12.99 812 0.5318 0.8506 0.8535 0.8506 0.8493
0.3717 14.0 875 0.5059 0.8579 0.8582 0.8579 0.8574
0.3343 14.99 937 0.5235 0.8472 0.8492 0.8472 0.8474
0.3053 16.0 1000 0.5226 0.8591 0.8571 0.8591 0.8567
0.2607 16.99 1062 0.5654 0.8591 0.8579 0.8591 0.8572
0.2814 18.0 1125 0.5622 0.8546 0.8541 0.8546 0.8537
0.2735 18.99 1187 0.6185 0.8506 0.8525 0.8506 0.8508
0.2673 20.0 1250 0.6210 0.8574 0.8544 0.8574 0.8550
0.2595 20.99 1312 0.6334 0.8422 0.8415 0.8422 0.8399
0.2583 22.0 1375 0.6565 0.8540 0.8545 0.8540 0.8527
0.239 22.99 1437 0.6859 0.8455 0.8458 0.8455 0.8447
0.2174 24.0 1500 0.6709 0.8591 0.8581 0.8591 0.8581
0.2288 24.99 1562 0.7437 0.8444 0.8426 0.8444 0.8419
0.2305 26.0 1625 0.7048 0.8529 0.8497 0.8529 0.8505
0.2071 26.99 1687 0.7152 0.8540 0.8527 0.8540 0.8529
0.2282 28.0 1750 0.7273 0.8568 0.8559 0.8568 0.8554
0.209 28.99 1812 0.7213 0.8557 0.8534 0.8557 0.8540
0.2078 29.76 1860 0.7273 0.8563 0.8544 0.8563 0.8548

Framework versions

  • Transformers 4.37.2
  • Pytorch 2.3.0
  • Datasets 2.19.1
  • Tokenizers 0.15.1
Downloads last month
0
Safetensors
Model size
195M params
Tensor type
F32
·
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 matthieulel/swinv2-large-patch4-window12to16-192to256-22kto1k-ft-finetuned-galaxy10-decals