implement onnx export for inception3/4, resnext, mobilenetv2 (#346)

* add inceptionv4 backbone/training settings
* add converted backbone, top-1 acc 80.08

Author: lostkevin

configs/classification/imagenet/inception/inceptionv3_b32x8_100e.py

@@ -4,16 +4,31 @@
 # model settings
 model = dict(
     type='Classification',
-    backbone=dict(type='Inception3'),
-    head=dict(
-        type='ClsHead',
-        with_avg_pool=True,
-        in_channels=2048,
-        loss_config=dict(
-            type='CrossEntropyLossWithLabelSmooth',
-            label_smooth=0,
+    backbone=dict(type='Inception3', num_classes=1000),
+    head=[
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=2048,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[1],
         ),
-        num_classes=num_classes))
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=768,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[0],
+        )
+    ])
 
 class_list = [
     '0', '1', '2', '3', '4', '5', '6', '7', '8', '9', '10', '11', '12', '13',
@@ -196,3 +211,5 @@
     interval=10,
     hooks=[dict(type='TextLoggerHook'),
            dict(type='TensorboardLoggerHook')])
+
+export = dict(export_type='raw', export_neck=True)

configs/classification/imagenet/inception/inceptionv4_b32x8_100e.py

@@ -0,0 +1,33 @@
+_base_ = 'configs/classification/imagenet/inception/inceptionv3_b32x8_100e.py'
+
+num_classes = 1000
+# model settings
+model = dict(
+    type='Classification',
+    backbone=dict(type='Inception4', num_classes=num_classes),
+    head=[
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=1536,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[1],
+        ),
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=768,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[0],
+        )
+    ])
+
+img_norm_cfg = dict(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])

configs/classification/imagenet/inception/inceptionv4_b32x8_200e_rmsprop.py

@@ -0,0 +1,46 @@
+# A config with the optimization settings from https://arxiv.org/pdf/1602.07261
+# May run with 20 GPUs
+_base_ = 'configs/classification/imagenet/inception/inceptionv3_b32x8_100e.py'
+
+num_classes = 1000
+# model settings
+model = dict(
+    type='Classification',
+    backbone=dict(type='Inception4', num_classes=num_classes),
+    head=[
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=1536,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[1],
+        ),
+        dict(
+            type='ClsHead',
+            with_fc=False,
+            in_channels=768,
+            loss_config=dict(
+                type='CrossEntropyLossWithLabelSmooth',
+                label_smooth=0,
+            ),
+            num_classes=num_classes,
+            input_feature_index=[0],
+        )
+    ])
+
+img_norm_cfg = dict(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+# optimizer
+optimizer = dict(
+    type='RMSprop', lr=0.045, momentum=0.9, weight_decay=0.9, eps=1.0)
+
+# learning policy
+lr_config = dict(policy='exp', gamma=0.96954)  # gamma**2 ~ 0.94
+checkpoint_config = dict(interval=10)
+
+# runtime settings
+total_epochs = 200

configs/classification/imagenet/mobilenet/mobilenetv2.py

@@ -13,7 +13,8 @@
             type='CrossEntropyLossWithLabelSmooth',
             label_smooth=0,
         ),
-        num_classes=num_classes))
+        num_classes=num_classes),
+    pretrained=True)
 
 # optimizer
 optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
@@ -25,4 +26,4 @@
 # runtime settings
 total_epochs = 100
 checkpoint_sync_export = True
-export = dict(export_neck=True)
+export = dict(export_type='raw', export_neck=True)

configs/classification/imagenet/resnext/resnext50-32x4d_b32x8_100e_jpg.py

@@ -19,7 +19,8 @@
             type='CrossEntropyLossWithLabelSmooth',
             label_smooth=0,
         ),
-        num_classes=num_classes))
+        num_classes=num_classes),
+    pretrained=True)
 
 # optimizer
 optimizer = dict(type='SGD', lr=0.1, momentum=0.9, weight_decay=0.0001)
@@ -30,3 +31,4 @@
 
 # runtime settings
 total_epochs = 100
+export = dict(export_type='raw', export_neck=True)

easycv/apis/export.py

@@ -158,34 +158,49 @@ def _get_blade_model():
 
 
 def _export_onnx_cls(model, model_config, cfg, filename, meta):
+    support_backbones = {
+        'ResNet': {
+            'depth': [50]
+        },
+        'MobileNetV2': {},
+        'Inception3': {},
+        'Inception4': {},
+        'ResNeXt': {
+            'depth': [50]
+        }
+    }
+    if model_config['backbone'].get('type', None) not in support_backbones:
+        tmp = ' '.join(support_backbones.keys())
+        info_str = f'Only support export onnx model for {tmp} now!'
+        raise ValueError(info_str)
+    configs = support_backbones[model_config['backbone'].get('type')]
+    for k, v in configs.items():
+        if v[0].__class__(model_config['backbone'].get(k, None)) not in v:
+            raise ValueError(
+                f"Unsupport config for {model_config['backbone'].get('type')}")
+
+    # save json config for test_pipline and class
+    with io.open(
+            filename +
+            '.config.json' if filename.endswith('onnx') else filename +
+            '.onnx.config.json', 'w') as ofile:
+        json.dump(meta, ofile)
 
-    if model_config['backbone'].get(
-            'type', None) == 'ResNet' and model_config['backbone'].get(
-                'depth', None) == 50:
-        # save json config for test_pipline and class
-        with io.open(
-                filename +
-                '.config.json' if filename.endswith('onnx') else filename +
-                '.onnx.config.json', 'w') as ofile:
-            json.dump(meta, ofile)
-
-        device = 'cuda' if torch.cuda.is_available() else 'cpu'
-        model.eval()
-        model.to(device)
-        img_size = int(cfg.image_size2)
-        x_input = torch.randn((1, 3, img_size, img_size)).to(device)
-        torch.onnx.export(
-            model,
-            (x_input, 'onnx'),
-            filename if filename.endswith('onnx') else filename + '.onnx',
-            export_params=True,
-            opset_version=12,
-            do_constant_folding=True,
-            input_names=['input'],
-            output_names=['output'],
-        )
-    else:
-        raise ValueError('Only support export onnx model for ResNet now!')
+    device = 'cuda' if torch.cuda.is_available() else 'cpu'
+    model.eval()
+    model.to(device)
+    img_size = int(cfg.image_size2)
+    x_input = torch.randn((1, 3, img_size, img_size)).to(device)
+    torch.onnx.export(
+        model,
+        (x_input, 'onnx'),
+        filename if filename.endswith('onnx') else filename + '.onnx',
+        export_params=True,
+        opset_version=12,
+        do_constant_folding=True,
+        input_names=['input'],
+        output_names=['output'],
+    )
 
 
 def _export_cls(model, cfg, filename):

easycv/models/backbones/__init__.py

@@ -10,6 +10,7 @@
 from .genet import PlainNet
 from .hrnet import HRNet
 from .inceptionv3 import Inception3
+from .inceptionv4 import Inception4
 from .lighthrnet import LiteHRNet
 from .mae_vit_transformer import *
 from .mit import MixVisionTransformer

easycv/models/backbones/inceptionv3.py

@@ -2,9 +2,6 @@
 r""" This model is taken from the official PyTorch model zoo.
      - torchvision.models.inception.py on 31th Aug, 2019
 """
-
-from collections import namedtuple
-
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -16,8 +13,6 @@
 
 __all__ = ['Inception3']
 
-_InceptionOutputs = namedtuple('InceptionOutputs', ['logits', 'aux_logits'])
-
 
 @BACKBONES.register_module
 class Inception3(nn.Module):
@@ -113,6 +108,7 @@ def forward(self, x):
         # N x 768 x 17 x 17
         x = self.Mixed_6e(x)
         # N x 768 x 17 x 17
+        aux = None
         if self.training and self.aux_logits:
             aux = self.AuxLogits(x)
         # N x 768 x 17 x 17
@@ -132,10 +128,7 @@ def forward(self, x):
         if hasattr(self, 'fc'):
             x = self.fc(x)
 
-        # N x 1000 (num_classes)
-        if self.training and self.aux_logits and hasattr(self, 'fc'):
-            return [_InceptionOutputs(x, aux)]
-        return [x]
+        return [aux, x]
 
 
 class InceptionA(nn.Module):