add url to pretrained checkpoints to README.md, small corrections

thomasverelst · Jun 18, 2020 · a29b33f · a29b33f
1 parent 63e086e
commit a29b33f
Show file tree

Hide file tree

Showing 5 changed files with 26 additions and 8 deletions.
diff --git a/classification/README.md b/classification/README.md
@@ -17,7 +17,19 @@ The other requirements (matplotlib, tqdm, numpy...) can be installed using pip:
 
     pip install -r requirements.txt
 
-### Evaluate a pretrained sparse DynConv network
+### Trained models
+Our trained models can be found here: [Microsoft OneDrive link](https://1drv.ms/u/s!ApImBF1PK3gnjoocGrSm908HR9-xuw?e=BWkRZF)
+
+Unzip and place them into the root of the classification folder, so the folder structure looks as follows:
+> ./main_cifar.py  
+> ./main_imagenet.py  
+> ./exp/cifar/resnet32/...  
+> ./exp/cifar/resnet26/...  
+> ./exp/imagenet/...  
+> ...  
+
+
+### Evaluate a trained sparse DynConv network
 
     python main_cifar.py --model resnet32 -r exp/cifar/resnet32/sparse07/checkpoint_best.pth --budget 0.7 -e
 

diff --git a/classification/main_cifar.py b/classification/main_cifar.py
@@ -223,7 +223,7 @@ def validate(args, val_loader, model, criterion, epoch):
                 plt.show()
 
     print(f'* Epoch {epoch} - Prec@1 {top1.avg:.3f}')
-    print(f'* FLOPS (multiply-accumulates, MACs) per image:  {model.compute_average_flops_cost()[0]/1e6:.6f} MMac')
+    print(f'* average FLOPS (multiply-accumulates, MACs) per image:  {model.compute_average_flops_cost()[0]/1e6:.6f} MMac')
     model.stop_flops_count()
     return top1.avg
 

diff --git a/classification/main_imagenet.py b/classification/main_imagenet.py
@@ -229,7 +229,7 @@ def validate(args, val_loader, model, criterion, epoch):
                 viz.showKey()
 
     print(f'* Epoch {epoch} - Prec@1 {top1.avg:.3f}')
-    print(f'* FLOPS (multiply-accumulates, MACs) per image:  {model.compute_average_flops_cost()[0]/1e6:.6f} MMac')
+    print(f'* average FLOPS (multiply-accumulates, MACs) per image:  {model.compute_average_flops_cost()[0]/1e6:.6f} MMac')
     model.stop_flops_count()
     return top1.avg
 

diff --git a/classification/models/resnet_util.py b/classification/models/resnet_util.py
@@ -22,8 +22,14 @@ class BasicBlock(nn.Module):
     """Standard residual block """
     expansion = 1
 
-    def __init__(self, inplanes, planes, stride=1, downsample=None, sparse=False):
+    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
+                 base_width=64, dilation=1, norm_layer=None, sparse=False):
         super(BasicBlock, self).__init__()
+        assert groups == 1
+        assert dilation == 1
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+
         self.conv1 = conv3x3(inplanes, planes, stride)
         self.bn1 = nn.BatchNorm2d(planes)
         self.relu = nn.ReLU(inplace=True)

diff --git a/pose/lib/core/function.py b/pose/lib/core/function.py
@@ -327,7 +327,7 @@ def validate(config, val_loader, val_dataset, model, criterion, output_dir,
 
     avg_flops, total_flops, batch_count = model.compute_average_flops_cost()
     logger.info(f'# PARAMS: {get_model_parameters_number(model, as_string=False)/1e6} M')
-    logger.info(f'# FLOPS (multiply-accumulates, MACs): {(total_flops/idx)/1e9} GMacs on {idx} images')
+    logger.info(f'# average FLOPS (multiply-accumulates, MACs) per image: {(total_flops/idx)/1e9} GMacs on {idx} images')
 
     # some conditional execution statistics
     if len(flops_per_layer) > 0:
@@ -342,17 +342,17 @@ def validate(config, val_loader, val_dataset, model, criterion, output_dir,
         s = ''
         for perc in perc_per_layer_avg:
             s += f'{round(float(perc), 2)}, '
-        logger.info(f'# FLOPS (multiply-accumulates MACs) used percentage per layer (average): {s}')
+        logger.info(f'# average FLOPS (multiply-accumulates MACs) used percentage per layer (average): {s}')
 
         s = ''
         for std in perc_per_layer_std:
             s += f'{round(float(std), 2)}, '
-        logger.info(f'# FLOPS (multiply-accumulates MACs) used percentage per layer (standard deviation): {s}')
+        logger.info(f'# average FLOPS (multiply-accumulates MACs) used percentage per layer (standard deviation): {s}')
 
 
         exec_cond_flops = int(torch.sum(flops_per_layer))/idx
         total_cond_flops = int(torch.sum(total_per_layer))/idx
-        logger.info(f'# Conditional FLOPS (multiply-accumulates MACs) over all layers (average per image): {exec_cond_flops/1e9} GMac out of {total_cond_flops/1e9} GMac ({round(100*exec_cond_flops/total_cond_flops,1)}%)')
+        logger.info(f'# Conditional average FLOPS (multiply-accumulates MACs) over all layers (average per image): {exec_cond_flops/1e9} GMac out of {total_cond_flops/1e9} GMac ({round(100*exec_cond_flops/total_cond_flops,1)}%)')
 
     return perf_indicator