from torchgeo.datasets import EuroSAT
from torchgeo.datamodules import EuroSATDataModule
from torchgeo.models import ResNet50_Weights, resnet50
import fastai.vision.all as fv
from fastai_torchgeo.data import GeoImageBlock
from fastai_torchgeo.resnet import make_resnet_model, resnet_splitEuroSAT Land Use/Land Cover (LULC) Classification Tutorial
finetune a torchgeo pretrained resnet model on the torchgeo EuroSAT dataset for LULC
This is tutorial on finetuning a pretrained torchgeo resnet model on the torchgeo EuroSAT dataset using the fastai framework.
Note: this tutorial assumes some familiarity with the fastai deep learning package and will focus on torchgeo integration.
Installation
Install the package
pip install git+https://github.com/butchland/fastai-torchgeo.gitImport the packages and download the EuroSAT dataset
Create a fastai datablock and fastai dataloaders
batch_size=64
num_workers = fv.defaults.cpus
dset_name = 'EuroSAT'dblock = fv.DataBlock(blocks=(GeoImageBlock(), fv.CategoryBlock()),
get_items=fv.get_image_files,
splitter=fv.RandomSplitter(valid_pct=0.2, seed=42),
get_y=fv.parent_label,
item_tfms=fv.Resize(64),
batch_tfms=[fv.Normalize.from_stats(EuroSATDataModule.mean, EuroSATDataModule.std)],
)cfg = fv.fastai_cfg()
data_dir = cfg.path('data')
dset_path = data_dir/dset_namedatamodule = EuroSATDataModule(root=dset_path,batch_size=batch_size, num_workers=num_workers, download=True)datamodule.prepare_data()CPU times: user 72.8 ms, sys: 8.16 ms, total: 80.9 ms
Wall time: 80.2 msdblock.summary(dset_path, show_batch=True)Setting-up type transforms pipelines
Collecting items from /home/studio-lab-user/.fastai/data/EuroSAT
Found 27000 items
2 datasets of sizes 21600,5400
Setting up Pipeline: partial
Setting up Pipeline: parent_label -> Categorize -- {'vocab': None, 'sort': True, 'add_na': False}
Building one sample
Pipeline: partial
starting from
/home/studio-lab-user/.fastai/data/EuroSAT/ds/images/remote_sensing/otherDatasets/sentinel_2/tif/Highway/Highway_427.tif
applying partial gives
GeoTensorImage of size 13x64x64
Pipeline: parent_label -> Categorize -- {'vocab': None, 'sort': True, 'add_na': False}
starting from
/home/studio-lab-user/.fastai/data/EuroSAT/ds/images/remote_sensing/otherDatasets/sentinel_2/tif/Highway/Highway_427.tif
applying parent_label gives
Highway
applying Categorize -- {'vocab': None, 'sort': True, 'add_na': False} gives
TensorCategory(3)
Final sample: (GeoTensorImage: torch.Size([13, 64, 64]), TensorCategory(3))
Collecting items from /home/studio-lab-user/.fastai/data/EuroSAT
Found 27000 items
2 datasets of sizes 21600,5400
Setting up Pipeline: partial
Setting up Pipeline: parent_label -> Categorize -- {'vocab': None, 'sort': True, 'add_na': False}
Setting up after_item: Pipeline: Resize -- {'size': (64, 64), 'method': 'crop', 'pad_mode': 'reflection', 'resamples': (<Resampling.BILINEAR: 2>, <Resampling.NEAREST: 0>), 'p': 1.0} -> ToTensor
Setting up before_batch: Pipeline:
Setting up after_batch: Pipeline: Normalize -- {'mean': tensor([[[[1354.4055]],
[[1118.2440]],
[[1042.9298]],
[[ 947.6262]],
[[1199.4729]],
[[1999.7909]],
[[2369.2229]],
[[2296.8262]],
[[ 732.0834]],
[[ 12.1133]],
[[1819.0103]],
[[1118.9240]],
[[2594.1409]]]], device='cuda:0'), 'std': tensor([[[[ 245.7176]],
[[ 333.0078]],
[[ 395.0925]],
[[ 593.7505]],
[[ 566.4170]],
[[ 861.1840]],
[[1086.6313]],
[[1117.9817]],
[[ 404.9198]],
[[ 4.7758]],
[[1002.5877]],
[[ 761.3032]],
[[1231.5858]]]], device='cuda:0'), 'axes': (0, 2, 3)}
Building one batch
Applying item_tfms to the first sample:
Pipeline: Resize -- {'size': (64, 64), 'method': 'crop', 'pad_mode': 'reflection', 'resamples': (<Resampling.BILINEAR: 2>, <Resampling.NEAREST: 0>), 'p': 1.0} -> ToTensor
starting from
(GeoTensorImage of size 13x64x64, TensorCategory(3))
applying Resize -- {'size': (64, 64), 'method': 'crop', 'pad_mode': 'reflection', 'resamples': (<Resampling.BILINEAR: 2>, <Resampling.NEAREST: 0>), 'p': 1.0} gives
(GeoTensorImage of size 13x64x64, TensorCategory(3))
applying ToTensor gives
(GeoTensorImage of size 13x64x64, TensorCategory(3))
Adding the next 3 samples
No before_batch transform to apply
Collating items in a batch
Applying batch_tfms to the batch built
Pipeline: Normalize -- {'mean': tensor([[[[1354.4055]],
[[1118.2440]],
[[1042.9298]],
[[ 947.6262]],
[[1199.4729]],
[[1999.7909]],
[[2369.2229]],
[[2296.8262]],
[[ 732.0834]],
[[ 12.1133]],
[[1819.0103]],
[[1118.9240]],
[[2594.1409]]]], device='cuda:0'), 'std': tensor([[[[ 245.7176]],
[[ 333.0078]],
[[ 395.0925]],
[[ 593.7505]],
[[ 566.4170]],
[[ 861.1840]],
[[1086.6313]],
[[1117.9817]],
[[ 404.9198]],
[[ 4.7758]],
[[1002.5877]],
[[ 761.3032]],
[[1231.5858]]]], device='cuda:0'), 'axes': (0, 2, 3)}
starting from
(GeoTensorImage of size 4x13x64x64, TensorCategory([3, 0, 7, 1], device='cuda:0'))
applying Normalize -- {'mean': tensor([[[[1354.4055]],
[[1118.2440]],
[[1042.9298]],
[[ 947.6262]],
[[1199.4729]],
[[1999.7909]],
[[2369.2229]],
[[2296.8262]],
[[ 732.0834]],
[[ 12.1133]],
[[1819.0103]],
[[1118.9240]],
[[2594.1409]]]], device='cuda:0'), 'std': tensor([[[[ 245.7176]],
[[ 333.0078]],
[[ 395.0925]],
[[ 593.7505]],
[[ 566.4170]],
[[ 861.1840]],
[[1086.6313]],
[[1117.9817]],
[[ 404.9198]],
[[ 4.7758]],
[[1002.5877]],
[[ 761.3032]],
[[1231.5858]]]], device='cuda:0'), 'axes': (0, 2, 3)} gives
(GeoTensorImage of size 4x13x64x64, TensorCategory([3, 0, 7, 1], device='cuda:0'))
dls = dblock.dataloaders(dset_path, bs=batch_size)dls.show_batch()
Download the torchgeo pretrained resnet model and prepare a fastai compatible model
pretrained = resnet50(ResNet50_Weights.SENTINEL2_ALL_MOCO, num_classes=10) # load pretrained weightsmodel = make_resnet_model(pretrained, n_out=10)Create a fastai learner
learn = fv.Learner(
dls,
model,
loss_func=fv.CrossEntropyLossFlat(),
metrics=[fv.accuracy],
splitter=resnet_split,
)
# freeze uses parameter groups created by `resnet_split`
# to lock parameters of pretrained model except for the model head
learn.freeze()# note: only head parameter group is trainable (except BatchNorm layers w/ch are always trainable)
learn.summary()Sequential (Input shape: 64 x 13 x 64 x 64)
============================================================================
Layer (type) Output Shape Param # Trainable
============================================================================
64 x 64 x 32 x 32
Conv2d 40768 False
BatchNorm2d 128 True
ReLU
____________________________________________________________________________
64 x 64 x 16 x 16
MaxPool2d
Conv2d 4096 False
BatchNorm2d 128 True
ReLU
Conv2d 36864 False
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 False
BatchNorm2d 512 True
ReLU
Conv2d 16384 False
BatchNorm2d 512 True
____________________________________________________________________________
64 x 64 x 16 x 16
Conv2d 16384 False
BatchNorm2d 128 True
ReLU
Conv2d 36864 False
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 False
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 64 x 16 x 16
Conv2d 16384 False
BatchNorm2d 128 True
ReLU
Conv2d 36864 False
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 False
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 128 x 16 x 16
Conv2d 32768 False
BatchNorm2d 256 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 147456 False
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 False
BatchNorm2d 1024 True
ReLU
Conv2d 131072 False
BatchNorm2d 1024 True
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 False
BatchNorm2d 256 True
ReLU
Conv2d 147456 False
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 False
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 False
BatchNorm2d 256 True
ReLU
Conv2d 147456 False
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 False
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 False
BatchNorm2d 256 True
ReLU
Conv2d 147456 False
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 False
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 256 x 8 x 8
Conv2d 131072 False
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
Conv2d 524288 False
BatchNorm2d 2048 True
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 False
BatchNorm2d 512 True
ReLU
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 False
BatchNorm2d 512 True
ReLU
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 False
BatchNorm2d 512 True
ReLU
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 False
BatchNorm2d 512 True
ReLU
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 False
BatchNorm2d 512 True
ReLU
Conv2d 589824 False
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 False
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 512 x 4 x 4
Conv2d 524288 False
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 2359296 False
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 False
BatchNorm2d 4096 True
ReLU
Conv2d 2097152 False
BatchNorm2d 4096 True
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 1048576 False
BatchNorm2d 1024 True
ReLU
Conv2d 2359296 False
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 False
BatchNorm2d 4096 True
ReLU
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 1048576 False
BatchNorm2d 1024 True
ReLU
Conv2d 2359296 False
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 False
BatchNorm2d 4096 True
ReLU
____________________________________________________________________________
64 x 2048 x 1 x 1
AdaptiveAvgPool2d
AdaptiveMaxPool2d
____________________________________________________________________________
64 x 4096
Flatten
BatchNorm1d 8192 True
Dropout
____________________________________________________________________________
64 x 512
Linear 2097152 True
ReLU
BatchNorm1d 1024 True
Dropout
____________________________________________________________________________
64 x 10
Linear 5120 True
____________________________________________________________________________
Total params: 25,650,880
Total trainable params: 2,164,608
Total non-trainable params: 23,486,272
Optimizer used: <function Adam>
Loss function: FlattenedLoss of CrossEntropyLoss()
Model frozen up to parameter group #2
Callbacks:
- TrainEvalCallback
- CastToTensor
- Recorder
- ProgressCallbackTrain the model
learn.lr_find()SuggestedLRs(valley=0.0006918309954926372)
from fastai.callback.tracker import SaveModelCallback
import numpy as nplearn.fine_tune(10, freeze_epochs=3,base_lr=7e-4, cbs=[SaveModelCallback(monitor='accuracy',fname='euronet-resnet50-stage1')])| epoch | train_loss | valid_loss | accuracy | time |
|---|---|---|---|---|
| 0 | 0.494235 | 0.293842 | 0.907778 | 00:41 |
| 1 | 0.333531 | 0.231448 | 0.921481 | 00:41 |
| 2 | 0.239049 | 0.184895 | 0.939444 | 00:41 |
Better model found at epoch 0 with accuracy value: 0.9077777862548828.
Better model found at epoch 1 with accuracy value: 0.9214814901351929.
Better model found at epoch 2 with accuracy value: 0.9394444227218628.
Better model found at epoch 0 with accuracy value: 0.959074079990387.
Better model found at epoch 1 with accuracy value: 0.9631481766700745.
Better model found at epoch 3 with accuracy value: 0.9688888788223267.
Better model found at epoch 4 with accuracy value: 0.9707407355308533.
Better model found at epoch 5 with accuracy value: 0.9722222089767456.
Better model found at epoch 6 with accuracy value: 0.9733333587646484.
Better model found at epoch 7 with accuracy value: 0.9740740656852722.
Better model found at epoch 9 with accuracy value: 0.9762963056564331.| epoch | train_loss | valid_loss | accuracy | time |
|---|---|---|---|---|
| 0 | 0.162829 | 0.124671 | 0.959074 | 00:43 |
| 1 | 0.088713 | 0.112167 | 0.963148 | 00:43 |
| 2 | 0.063464 | 0.126567 | 0.962037 | 00:43 |
| 3 | 0.055648 | 0.104110 | 0.968889 | 00:43 |
| 4 | 0.032861 | 0.103171 | 0.970741 | 00:45 |
| 5 | 0.021249 | 0.102579 | 0.972222 | 00:44 |
| 6 | 0.010938 | 0.102371 | 0.973333 | 00:43 |
| 7 | 0.005382 | 0.097377 | 0.974074 | 00:43 |
| 8 | 0.004003 | 0.101503 | 0.973889 | 00:44 |
| 9 | 0.002489 | 0.098339 | 0.976296 | 00:44 |
learn.save('euronet-resnet50-stage1-final')Path('models/euronet-resnet50-stage1-final.pth')learn.recorder.plot_loss()
learn.summary()Sequential (Input shape: 64 x 13 x 64 x 64)
============================================================================
Layer (type) Output Shape Param # Trainable
============================================================================
64 x 64 x 32 x 32
Conv2d 40768 True
BatchNorm2d 128 True
ReLU
____________________________________________________________________________
64 x 64 x 16 x 16
MaxPool2d
Conv2d 4096 True
BatchNorm2d 128 True
ReLU
Conv2d 36864 True
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 True
BatchNorm2d 512 True
ReLU
Conv2d 16384 True
BatchNorm2d 512 True
____________________________________________________________________________
64 x 64 x 16 x 16
Conv2d 16384 True
BatchNorm2d 128 True
ReLU
Conv2d 36864 True
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 True
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 64 x 16 x 16
Conv2d 16384 True
BatchNorm2d 128 True
ReLU
Conv2d 36864 True
BatchNorm2d 128 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 256 x 16 x 16
Conv2d 16384 True
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 128 x 16 x 16
Conv2d 32768 True
BatchNorm2d 256 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 147456 True
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 True
BatchNorm2d 1024 True
ReLU
Conv2d 131072 True
BatchNorm2d 1024 True
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 True
BatchNorm2d 256 True
ReLU
Conv2d 147456 True
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 True
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 True
BatchNorm2d 256 True
ReLU
Conv2d 147456 True
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 True
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 128 x 8 x 8
Conv2d 65536 True
BatchNorm2d 256 True
ReLU
Conv2d 147456 True
BatchNorm2d 256 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 512 x 8 x 8
Conv2d 65536 True
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 256 x 8 x 8
Conv2d 131072 True
BatchNorm2d 512 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
Conv2d 524288 True
BatchNorm2d 2048 True
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 True
BatchNorm2d 512 True
ReLU
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 True
BatchNorm2d 512 True
ReLU
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 True
BatchNorm2d 512 True
ReLU
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 True
BatchNorm2d 512 True
ReLU
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 256 x 4 x 4
Conv2d 262144 True
BatchNorm2d 512 True
ReLU
Conv2d 589824 True
BatchNorm2d 512 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 1024 x 4 x 4
Conv2d 262144 True
BatchNorm2d 2048 True
ReLU
____________________________________________________________________________
64 x 512 x 4 x 4
Conv2d 524288 True
BatchNorm2d 1024 True
ReLU
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 2359296 True
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 True
BatchNorm2d 4096 True
ReLU
Conv2d 2097152 True
BatchNorm2d 4096 True
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 1048576 True
BatchNorm2d 1024 True
ReLU
Conv2d 2359296 True
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 True
BatchNorm2d 4096 True
ReLU
____________________________________________________________________________
64 x 512 x 2 x 2
Conv2d 1048576 True
BatchNorm2d 1024 True
ReLU
Conv2d 2359296 True
BatchNorm2d 1024 True
Identity
ReLU
Identity
____________________________________________________________________________
64 x 2048 x 2 x 2
Conv2d 1048576 True
BatchNorm2d 4096 True
ReLU
____________________________________________________________________________
64 x 2048 x 1 x 1
AdaptiveAvgPool2d
AdaptiveMaxPool2d
____________________________________________________________________________
64 x 4096
Flatten
BatchNorm1d 8192 True
Dropout
____________________________________________________________________________
64 x 512
Linear 2097152 True
ReLU
BatchNorm1d 1024 True
Dropout
____________________________________________________________________________
64 x 10
Linear 5120 True
____________________________________________________________________________
Total params: 25,650,880
Total trainable params: 25,650,880
Total non-trainable params: 0
Optimizer used: <function Adam>
Loss function: FlattenedLoss of CrossEntropyLoss()
Model unfrozen
Callbacks:
- TrainEvalCallback
- CastToTensor
- Recorder
- ProgressCallbacklearn.fit_one_cycle(20, lr_max=slice(3e-3, 6e-6), cbs=[SaveModelCallback(monitor='accuracy',fname='euronet-resnet50-stage2')])| epoch | train_loss | valid_loss | accuracy | time |
|---|---|---|---|---|
| 0 | 0.035239 | 0.170115 | 0.959259 | 00:44 |
| 1 | 0.154670 | 0.230367 | 0.944815 | 00:44 |
| 2 | 0.207444 | 0.338804 | 0.908889 | 00:44 |
| 3 | 0.170779 | 0.218512 | 0.938704 | 00:44 |
| 4 | 0.176847 | 0.216765 | 0.937593 | 00:44 |
| 5 | 0.159534 | 0.193209 | 0.940926 | 00:44 |
| 6 | 0.120023 | 0.179367 | 0.944444 | 00:44 |
| 7 | 0.095654 | 0.156686 | 0.953148 | 00:44 |
| 8 | 0.069015 | 0.181001 | 0.948519 | 00:44 |
| 9 | 0.053350 | 0.164088 | 0.952222 | 00:44 |
| 10 | 0.036882 | 0.124723 | 0.964630 | 00:43 |
| 11 | 0.021793 | 0.099748 | 0.972963 | 00:43 |
| 12 | 0.014808 | 0.104900 | 0.972593 | 00:44 |
| 13 | 0.009095 | 0.113750 | 0.973148 | 00:44 |
| 14 | 0.004226 | 0.097392 | 0.976111 | 00:44 |
| 15 | 0.003125 | 0.097975 | 0.976852 | 00:44 |
| 16 | 0.001740 | 0.090906 | 0.977593 | 00:44 |
| 17 | 0.002263 | 0.085452 | 0.979444 | 00:44 |
| 18 | 0.000893 | 0.086279 | 0.980185 | 00:44 |
| 19 | 0.000644 | 0.087188 | 0.979444 | 00:44 |
Better model found at epoch 0 with accuracy value: 0.9592592716217041.
Better model found at epoch 10 with accuracy value: 0.9646296501159668.
Better model found at epoch 11 with accuracy value: 0.9729629755020142.
Better model found at epoch 13 with accuracy value: 0.9731481671333313.
Better model found at epoch 14 with accuracy value: 0.976111114025116.
Better model found at epoch 15 with accuracy value: 0.9768518805503845.
Better model found at epoch 16 with accuracy value: 0.9775925874710083.
Better model found at epoch 17 with accuracy value: 0.9794444441795349.
Better model found at epoch 18 with accuracy value: 0.9801852107048035.Post training (error analysis and setting up for inference)
learn.recorder.plot_loss()
learn.save('euronet-resnet50-stage2-final')Path('models/euronet-resnet50-stage2-final.pth')learn.validate()(#2) [0.08627868443727493,0.9801852107048035]interp = fv.ClassificationInterpretation.from_learner(learn)interp.plot_confusion_matrix()
interp.plot_top_losses(9)