This branch is used to match test precision between openmmlab repo an…

…d HRNet raw repo;

* Add HRNet weights convert script (tools/scripts/convert_hrnet.py);

* Modify test_pipeline of pascal context dataset;

* Add ignore items;

.gitignore

@@ -108,6 +108,8 @@ data
 .idea
 
 # custom
+*.txt
+*.jpg
 *.npy
 *.pkl
 *.pkl.json

configs/_base_/datasets/pascal_context.py

@@ -23,7 +23,8 @@
     dict(type='LoadImageFromFile'),
     dict(
         type='MultiScaleFlipAug',
-        img_scale=img_scale,
+        img_scale=None,
+        img_ratios=[1.0],
         # img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
         flip=False,
         transforms=[

configs/_base_/datasets/pascal_context_59.py

@@ -23,7 +23,8 @@
     dict(type='LoadImageFromFile'),
     dict(
         type='MultiScaleFlipAug',
-        img_scale=img_scale,
+        img_scale=None,
+        img_ratios=[1.0],
         # img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
         flip=False,
         transforms=[

mmseg/datasets/pipelines/transforms.py

@@ -389,10 +389,16 @@ def __call__(self, results):
                 result dict.
         """
 
-        results['img'] = mmcv.imnormalize(results['img'], self.mean, self.std,
-                                          self.to_rgb)
-        results['img_norm_cfg'] = dict(
-            mean=self.mean, std=self.std, to_rgb=self.to_rgb)
+        # results['img'] = mmcv.imnormalize(results['img'], self.mean,
+        #                                   self.std, self.to_rgb)
+        # results['img_norm_cfg'] = dict(
+        #     mean=self.mean, std=self.std, to_rgb=self.to_rgb)
+        image = results['img']
+        image = image.astype(np.float32)[:, :, ::-1]
+        image = image / 255.0
+        image -= [0.485, 0.456, 0.406]
+        image /= [0.229, 0.224, 0.225]
+        results['img'] = image
         return results
 
     def __repr__(self):

mmseg/models/decode_heads/fcn_head.py

@@ -43,6 +43,7 @@ def __init__(self,
                 kernel_size=kernel_size,
                 padding=conv_padding,
                 dilation=dilation,
+                bias=True,
                 conv_cfg=self.conv_cfg,
                 norm_cfg=self.norm_cfg,
                 act_cfg=self.act_cfg))

tools/scripts/convert_hrnet.py

@@ -0,0 +1,44 @@
+import argparse
+from collections import OrderedDict
+
+import torch
+
+
+def convert(src, dst):
+    """Convert keys in detectron pretrained ResNet models to pytorch style."""
+    # convert to pytorch style
+    state_dict = OrderedDict()
+    src_dict = torch.load(src)
+    src_state_dict = src_dict.get('state_dict', src_dict)
+    for k, v in src_state_dict.items():
+        new_key = k.replace('model', 'backbone')
+        if new_key.startswith('backbone.last_layer.0'):
+            state_dict[new_key.replace('backbone.last_layer.0',
+                                       'decode_head.convs.0.conv')] = v
+        elif new_key.startswith('backbone.last_layer.1'):
+            state_dict[new_key.replace('backbone.last_layer.1',
+                                       'decode_head.convs.0.bn')] = v
+        elif new_key.startswith('backbone.last_layer.3'):
+            state_dict[new_key.replace('backbone.last_layer.3',
+                                       'decode_head.conv_seg')] = v
+        else:
+            state_dict[new_key] = v
+
+    # save checkpoint
+    checkpoint = dict()
+    checkpoint['state_dict'] = state_dict
+    assert len(state_dict) == len(src_state_dict)
+    checkpoint['meta'] = dict()
+    torch.save(checkpoint, dst)
+
+
+def main():
+    parser = argparse.ArgumentParser(description='Convert model keys')
+    parser.add_argument('src', help='src detectron model path')
+    parser.add_argument('dst', help='save path')
+    args = parser.parse_args()
+    convert(args.src, args.dst)
+
+
+if __name__ == '__main__':
+    main()

tools/test.py

@@ -117,9 +117,9 @@ def main():
     # build the model and load checkpoint
     cfg.model.train_cfg = None
     model = build_segmentor(cfg.model, test_cfg=cfg.get('test_cfg'))
-    checkpoint = load_checkpoint(model, args.checkpoint, map_location='cpu')
-    model.CLASSES = checkpoint['meta']['CLASSES']
-    model.PALETTE = checkpoint['meta']['PALETTE']
+    _ = load_checkpoint(model, args.checkpoint, map_location='cpu')
+    model.CLASSES = dataset.CLASSES
+    model.PALETTE = dataset.PALETTE
 
     efficient_test = False
     if args.eval_options is not None: