refactor: Removed contiguous param since it's included in torch>=1.7 (#…

…756)

* refactor: Removed contiguous params

* ci: Removed contiguous params from CI

* refactor: Refactored loaders usage in training script

* refactor: Removed unused import

.github/workflows/references.yml

@@ -60,7 +60,6 @@ jobs:
           python -m pip install --upgrade pip
           pip install -e .[torch] --upgrade
           pip install -r references/requirements.txt
-          pip install contiguous-params
           sudo apt-get update && sudo apt-get install fonts-freefont-ttf -y
       - if: matrix.framework == 'tensorflow'
         name: Train for a short epoch (TF)
@@ -121,7 +120,6 @@ jobs:
           python -m pip install --upgrade pip
           pip install -e .[torch] --upgrade
           pip install -r references/requirements.txt
-          pip install contiguous-params
       - name: Download and extract toy set
         run: |
           wget https://github.com/mindee/doctr/releases/download/v0.3.1/toy_recogition_set-036a4d80.zip
@@ -240,7 +238,6 @@ jobs:
           python -m pip install --upgrade pip
           pip install -e .[torch] --upgrade
           pip install -r references/requirements.txt
-          pip install contiguous-params
       - name: Download and extract toy set
         run: |
           wget https://github.com/mindee/doctr/releases/download/v0.3.1/toy_detection_set-bbbb4243.zip

references/classification/README.md

@@ -11,11 +11,6 @@ pip install -e . --upgrade
 pip install -r references/requirements.txt
 ```
 
-if you are using PyTorch back-end, there is an extra dependency (to optimize data loading):
-```shell
-pip install contiguous-params>=1.0.0
-```
-
 ## Usage
 
 You can start your training in TensorFlow:

references/classification/train_pytorch.py

@@ -15,7 +15,6 @@
 import numpy as np
 import torch
 import wandb
-from contiguous_params import ContiguousParams
 from fastprogress.fastprogress import master_bar, progress_bar
 from torch.nn.functional import cross_entropy
 from torch.optim.lr_scheduler import CosineAnnealingLR, MultiplicativeLR, OneCycleLR
@@ -105,10 +104,8 @@ def fit_one_epoch(model, train_loader, batch_transforms, optimizer, scheduler, m
         scaler = torch.cuda.amp.GradScaler()
 
     model.train()
-    train_iter = iter(train_loader)
     # Iterate over the batches of the dataset
-    for _ in progress_bar(range(len(train_loader)), parent=mb):
-        images, targets = next(train_iter)
+    for images, targets in progress_bar(train_loader, parent=mb):
 
         if torch.cuda.is_available():
             images = images.cuda()
@@ -141,8 +138,7 @@ def evaluate(model, val_loader, batch_transforms, amp=False):
     model.eval()
     # Validation loop
     val_loss, correct, samples, batch_cnt = 0, 0, 0, 0
-    val_iter = iter(val_loader)
-    for images, targets in val_iter:
+    for images, targets in val_loader:
         images = batch_transforms(images)
 
         if torch.cuda.is_available():
@@ -265,8 +261,7 @@ def main(args):
         return
 
     # Optimizer
-    model_params = ContiguousParams([p for p in model.parameters() if p.requires_grad]).contiguous()
-    optimizer = torch.optim.Adam(model_params, args.lr,
+    optimizer = torch.optim.Adam([p for p in model.parameters() if p.requires_grad], args.lr,
                                  betas=(0.95, 0.99), eps=1e-6, weight_decay=args.weight_decay)
 
     # LR Finder

references/detection/README.md

@@ -11,11 +11,6 @@ pip install -e . --upgrade
 pip install -r references/requirements.txt
 ```
 
-if you are using PyTorch back-end, there is an extra dependency (to optimize data loading):
-```shell
-pip install contiguous-params>=1.0.0
-```
-
 ## Usage
 
 You can start your training in TensorFlow:

references/detection/train_pytorch.py

@@ -16,7 +16,6 @@
 import numpy as np
 import torch
 import wandb
-from contiguous_params import ContiguousParams
 from fastprogress.fastprogress import master_bar, progress_bar
 from torch.optim.lr_scheduler import CosineAnnealingLR, MultiplicativeLR, OneCycleLR
 from torch.utils.data import DataLoader, RandomSampler, SequentialSampler
@@ -107,9 +106,8 @@ def fit_one_epoch(model, train_loader, batch_transforms, optimizer, scheduler, m
         scaler = torch.cuda.amp.GradScaler()
 
     model.train()
-    train_iter = iter(train_loader)
     # Iterate over the batches of the dataset
-    for images, targets in progress_bar(train_iter, parent=mb):
+    for images, targets in progress_bar(train_loader, parent=mb):
 
         if torch.cuda.is_available():
             images = images.cuda()
@@ -145,8 +143,7 @@ def evaluate(model, val_loader, batch_transforms, val_metric, amp=False):
     val_metric.reset()
     # Validation loop
     val_loss, batch_cnt = 0, 0
-    val_iter = iter(val_loader)
-    for images, targets in val_iter:
+    for images, targets in val_loader:
         if torch.cuda.is_available():
             images = images.cuda()
         images = batch_transforms(images)
@@ -272,8 +269,7 @@ def main(args):
             p.reguires_grad_(False)
 
     # Optimizer
-    model_params = ContiguousParams([p for p in model.parameters() if p.requires_grad]).contiguous()
-    optimizer = torch.optim.Adam(model_params, args.lr,
+    optimizer = torch.optim.Adam([p for p in model.parameters() if p.requires_grad], args.lr,
                                  betas=(0.95, 0.99), eps=1e-6, weight_decay=args.weight_decay)
     # LR Finder
     if args.find_lr:

references/obj_detection/train_pytorch.py

@@ -116,9 +116,8 @@ def fit_one_epoch(model, train_loader, optimizer, scheduler, mb, amp=False):
         scaler = torch.cuda.amp.GradScaler()
 
     model.train()
-    train_iter = iter(train_loader)
     # Iterate over the batches of the dataset
-    for images, targets in progress_bar(train_iter, parent=mb):
+    for images, targets in progress_bar(train_loader, parent=mb):
 
         targets = convert_to_abs_coords(targets, images.shape)
         if torch.cuda.is_available():
@@ -148,10 +147,7 @@ def fit_one_epoch(model, train_loader, optimizer, scheduler, mb, amp=False):
 def evaluate(model, val_loader, metric, amp=False):
     model.eval()
     metric.reset()
-    val_iter = iter(val_loader)
-    for images, targets in val_iter:
-
-        images, targets = next(val_iter)
+    for images, targets in val_loader:
         targets = convert_to_abs_coords(targets, images.shape)
         if torch.cuda.is_available():
             images = images.cuda()

references/recognition/README.md

@@ -11,11 +11,6 @@ pip install -e . --upgrade
 pip install -r references/requirements.txt
 ```
 
-if you are using PyTorch back-end, there is an extra dependency (to optimize data loading):
-```shell
-pip install contiguous-params>=1.0.0
-```
-
 ## Usage
 
 You can start your training in TensorFlow:

references/recognition/train_pytorch.py

@@ -17,7 +17,6 @@
 import numpy as np
 import torch
 import wandb
-from contiguous_params import ContiguousParams
 from fastprogress.fastprogress import master_bar, progress_bar
 from torch.optim.lr_scheduler import CosineAnnealingLR, MultiplicativeLR, OneCycleLR
 from torch.utils.data import DataLoader, RandomSampler, SequentialSampler
@@ -108,9 +107,8 @@ def fit_one_epoch(model, train_loader, batch_transforms, optimizer, scheduler, m
         scaler = torch.cuda.amp.GradScaler()
 
     model.train()
-    train_iter = iter(train_loader)
     # Iterate over the batches of the dataset
-    for images, targets in progress_bar(train_iter, parent=mb):
+    for images, targets in progress_bar(train_loader, parent=mb):
 
         if torch.cuda.is_available():
             images = images.cuda()
@@ -148,8 +146,7 @@ def evaluate(model, val_loader, batch_transforms, val_metric, amp=False):
     val_metric.reset()
     # Validation loop
     val_loss, batch_cnt = 0, 0
-    val_iter = iter(val_loader)
-    for images, targets in val_iter:
+    for images, targets in val_loader:
         if torch.cuda.is_available():
             images = images.cuda()
         images = batch_transforms(images)
@@ -279,8 +276,7 @@ def main(args):
         return
 
     # Optimizer
-    model_params = ContiguousParams([p for p in model.parameters() if p.requires_grad]).contiguous()
-    optimizer = torch.optim.Adam(model_params, args.lr,
+    optimizer = torch.optim.Adam([p for p in model.parameters() if p.requires_grad], args.lr,
                                  betas=(0.95, 0.99), eps=1e-6, weight_decay=args.weight_decay)
     # LR Finder
     if args.find_lr: