27bbc70d71
Add back old ModelEma and rename new one to ModelEmaV2 to avoid compat breaks in dependant code. Shuffle train script, add a few comments, remove DataParallel support, support experimental torchscript training.
4 years ago
9214ca0716
Simplifying EMA...
4 years ago
80078c47bb
Add Adafactor and Adahessian optimizers, cleanup optimizer arg passing, add gradient clipping support.
4 years ago
47a7b3b5b1
More flexible mixup mode, add 'half' mode.
4 years ago
532e3b417d
Reorg of utils into separate modules
4 years ago
751b0bba98
Add global_pool (--gp) arg changes to allow passing 'fast' easily for train/validate to avoid channels_last issue with AdaptiveAvgPool
4 years ago
9c297ec67d
Cleanup Apex vs native AMP scaler state save/load. Cleanup CheckpointSaver a bit.
4 years ago
c2cd1a332e
Improve torch amp support and add channels_last support for train/validate scripts
4 years ago
5f563ca4df
fix save_checkpoint bug with native amp
4 years ago
d98967ed5d
add support for native torch AMP in torch 1.6
4 years ago
8c9814e3f5
Final cleanup of mixup/cutmix. Element/batch modes working with both collate (prefetcher active) and without prefetcher.
4 years ago
f471c17c9d
More cutmix/mixup overhaul, ready to kick-off some trials.
4 years ago
92f2d0d65d
Merge branch 'master' into cutmix. Fixup a few issues.
4 years ago
fa28067704
Add more augmentation arguments, including a no_aug disable flag. Fix #209
4 years ago
7995295968
Merge branch 'logger' into features. Change 'logger' to '_logger'.
4 years ago
1998bd3180
Merge branch 'feature/AB/logger' of https://github.com/antoinebrl/pytorch-image-models into logger
4 years ago
6c17d57a2c
Fix some attributions, add copyrights to some file docstrings
4 years ago
78fa0772cc
Leverage python hierachical logger
...
with this update one can tune the kind of logs generated by timm but
training and inference traces are unchanged
5 years ago
6441e9cc1b
Fix memory_efficient mode for DenseNets. Add AntiAliasing (Blur) support for DenseNets and create one test model. Add lr cycle/mul params to train args.
5 years ago
a7f570c9b7
added MultiEpochsDataLoader
5 years ago
13cf68850b
Remove poorly named metrics from torch imagenet example origins. Use top1/top5 in csv output for consistency with existing validation results files, acc elsewhere. Fixes #111
5 years ago
27b3680d49
Revamp LR noise, move logic to scheduler base. Fixup PlateauLRScheduler and add it as an option.
5 years ago
514b0938c4
Experimenting with per-epoch learning rate noise w/ step scheduler
5 years ago
43225d110c
Unify drop connect vs drop path under 'drop path' name, switch all EfficientNet/MobilenetV3 refs to 'drop_path'. Update factory to handle new drop args.
5 years ago
b3cb5f3275
Working on CutMix impl as per #8 , integrating with Mixup, currently experimenting...
5 years ago
b72013def8
Added commandline argument validation-batch-size-multiplier with default set to 1.
5 years ago
5b7cc16ac9
Add warning about using sync-bn with zero initialized BN layers. Fixes #54
5 years ago
d9a6a9d0af
Merge pull request #74 from rwightman/augmix-jsd
...
AugMix, JSD loss, SplitBatchNorm (Auxiliary BN), and more
5 years ago
3eb4a96eda
Update AugMix, JSD, etc comments and references
5 years ago
7547119891
Add SplitBatchNorm. AugMix, Rand/AutoAugment, Split (Aux) BatchNorm, Jensen-Shannon Divergence, RandomErasing all working together
5 years ago
40fea63ebe
Add checkpoint averaging script. Add headers, shebangs, exec perms to all scripts
5 years ago
4666cc9aed
Add --pin-mem arg to enable dataloader pin_memory (showing more benefit in some scenarios now), also add --torchscript arg to validate.py for testing models with jit.script
5 years ago
232ab7fb12
Working on an implementation of AugMix with JensenShannonDivergence loss that's compatible with my AutoAugment and RandAugment impl
5 years ago
5719b493ad
Missed update dist-bn logic for EMA model
5 years ago
a435ea1327
Change reduce_bn to distribute_bn, add ability to choose between broadcast and reduce (mean). Add crop_pct arg to allow selecting validation crop while training.
5 years ago
3bff2b21dc
Add support for keeping running bn stats the same across distributed training nodes before eval/save
5 years ago
1f39d15f15
Allow float decay epochs arg for training, works out with step lr math
5 years ago
7b83e67f77
Pass drop connect arg through to EfficientNet models
5 years ago
4748c6dff2
Fix non-prefetch variant of Mixup. Fixes #50
5 years ago
187ecbafbe
Add support for loading args from yaml file (and saving them with each experiment)
5 years ago
b750b76f67
More AutoAugment work. Ready to roll...
5 years ago
3d9c8a6489
Add support for new AMP checkpointing support w/ amp.state_dict
5 years ago
fac58f609a
Add RAdam, NovoGrad, Lookahead, and AdamW optimizers, a few ResNet tweaks and scheduler factory tweak.
...
* Add some of the trendy new optimizers. Decent results but not clearly better than the standards.
* Can create a None scheduler for constant LR
* ResNet defaults to zero_init of last BN in residual
* add resnet50d config
5 years ago
66634d2200
Add support to split random erasing blocks into randomly selected number with --recount arg. Fix random selection of aspect ratios.
5 years ago
e7c8a37334
Make min-lr and cooldown-epochs cmdline args, change dash in color_jitter arg for consistency
5 years ago
c6b32cbe73
A number of tweaks to arguments, epoch handling, config
...
* reorganize train args
* allow resolve_data_config to be used with dict args, not just arparse
* stop incrementing epoch before save, more consistent naming vs csv, etc
* update resume and start epoch handling to match above
* stop auto-incrementing epoch in scheduler
5 years ago
b20bb58284
Distributed tweaks
...
* Support PyTorch native DDP as fallback if APEX not present
* Support SyncBN for both APEX and Torch native (if torch >= 1.1)
* EMA model does not appear to need DDP wrapper, no gradients, updated from wrapped model
5 years ago
6fc886acaf
Remove all prints, change most to logging calls, tweak alignment of batch logs, improve setup.py
5 years ago
aa4354f466
Big re-org, working towards making pip/module as 'timm'
5 years ago
7dab6d1ec7
Default to img_size in model default_cfg, defer output folder creation until later in the init sequence
6 years ago
9bcd65181b
Add exponential moving average for model weights + few other additions and cleanup
...
* ModelEma class added to track an EMA set of weights for the model being trained
* EMA handling added to train, validation and clean_checkpoint scripts
* Add multi checkpoint or multi-model validation support to validate.py
* Add syncbn option (APEX) to train script for experimentation
* Cleanup interface of CheckpointSaver while adding ema functionality
6 years ago
e6c14427c0
More appropriate/correct loss name
6 years ago
127487369f
Fix bug for prefetcher
...
Set input and target to Cuda when without using prefetcher.
6 years ago
4d2056722a
Mixup and prefetcher improvements
...
* Do mixup in custom collate fn if prefetcher enabled, reduces performance impact
* Move mixup code to own file
* Add arg to disable prefetcher
* Fix no cuda transfer when prefetcher off
* Random erasing when prefetcher off wasn't changed to match new args, fixed
* Default random erasing to off (prob = 0.) for train
6 years ago
780c0a96a4
Change args for RandomErasing so only one required for pixel/color mode
6 years ago
76539d905e
Some transform/data/loader refactoring, hopefully didn't break things
...
* factor out data related constants to own file
* move data related config helpers to own file
* add a variant of RandomResizeCrop that randomizes interpolation method
* remove old Numpy version of RandomErasing
* cleanup torch version of RandomErasing and use it in either GPU loader batch mode or single image cpu Transform
6 years ago
fee607edf6
Mixup implemention in progress
...
* initial impl w/ label smoothing converging, but needs more testing
6 years ago
8fbd62a169
Exclude batchnorm and bias params from weight_decay by default
6 years ago
bc264269c9
Morph mnasnet impl into a generic mobilenet that covers Mnasnet, MobileNetV1/V2, ChamNet, FBNet, and related
...
* add an alternate RMSprop opt that applies eps like TF
* add bn params for passing through alternates and changing defaults to TF style
6 years ago
e9c7961efc
Fix pooling in mnasnet, more sensible default for AMP opt level
6 years ago
0562b91c38
Add per model crop pct, interpolation defaults, tie it all together
...
* create one resolve fn to pull together model defaults + cmd line args
* update attribution comments in some models
* test update train/validation/inference scripts
6 years ago
c328b155e9
Random erasing crash fix and args pass through
6 years ago
9c3859fb9c
Uniform pretrained model handling.
...
* All models have 'default_cfgs' dict
* load/resume/pretrained helpers factored out
* pretrained load operates on state_dict based on default_cfg
* test all models in validate
* schedule, optim factor factored out
* test time pool wrapper applied based on default_cfg
6 years ago
f1cd1a5ce3
Cleanup CheckpointSaver, add support for increasing or decreasing metric, switch to prec1 metric in train loop
6 years ago
5180f94c7e
Distributed (multi-process) train, multi-gpu single process train, and NVIDIA AMP support
6 years ago
45cde6f0c7
Improve creation of data pipeline with prefetch enabled vs disabled, fixup inception_res_v2 and dpn models
6 years ago
2295cf56c2
Add some Nvidia performance enhancements (prefetch loader, fast collate), and refactor some of training and model fact/transforms
6 years ago
9d927a389a
Add adabound, random erasing
6 years ago
1577c52976
Resnext added, changes to bring it and seresnet in line with rest of models
6 years ago
31055466fc
Fixup validate/inference script args, fix senet init for better test accuracy
6 years ago
b1a5a71151
Update schedulers
6 years ago
b5255960d9
Tweaking tanh scheduler, senet weight init (for BN), transform defaults
6 years ago
a336e5bff3
Minor updates
6 years ago
cf0c280e1b
Cleanup tranforms, add custom schedulers, tweak senet34 model
6 years ago
c57717d325
Fix tta train bug, improve logging
6 years ago
72b4d162a2
Increase training performance
6 years ago
5855b07ae0
Initial commit, puting some ol pieces together
6 years ago
Ross Wightman