From 98a6b19112ae7de70bf9e75ee62eb75b995f70a0 Mon Sep 17 00:00:00 2001
From: still-wait <742925032@qq.com>
Date: Mon, 21 Sep 2020 12:06:04 +0000
Subject: [PATCH] fix train batch size

---
 configs/solov2/solov2_r50_fpn_3x.yml   | 90 ++++++++++++++++++++++++++
 ppdet/modeling/architectures/solov2.py | 16 +++--
 2 files changed, 100 insertions(+), 6 deletions(-)
 create mode 100644 configs/solov2/solov2_r50_fpn_3x.yml

diff --git a/configs/solov2/solov2_r50_fpn_3x.yml b/configs/solov2/solov2_r50_fpn_3x.yml
new file mode 100644
index 000000000..ee1b38eba
--- /dev/null
+++ b/configs/solov2/solov2_r50_fpn_3x.yml
@@ -0,0 +1,90 @@
+architecture: SOLOv2
+use_gpu: true
+max_iters: 270000
+snapshot_iter: 30000
+log_smooth_window: 20
+save_dir: output
+pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar
+metric: COCO
+weights: output/solov2/solov2_r50_fpn_3x/model_final
+num_classes: 81
+
+SOLOv2:
+  backbone: ResNet
+  fpn: FPN
+  bbox_head: SOLOv2Head
+  mask_head: SOLOv2MaskHead
+  batch_size: 2
+
+ResNet:
+  depth: 50
+  feature_maps: [2, 3, 4, 5]
+  freeze_at: 2
+  norm_type: bn
+
+FPN:
+  max_level: 6
+  min_level: 2
+  num_chan: 256
+  spatial_scale: [0.03125, 0.0625, 0.125, 0.25]
+  reverse_out: True
+
+SOLOv2Head:
+  seg_feat_channels: 512
+  stacked_convs: 4
+  num_grids: [40, 36, 24, 16, 12]
+  kernel_out_channels: 256
+
+SOLOv2MaskHead:
+  out_channels: 128
+  start_level: 0
+  end_level: 3
+  num_classes: 256
+
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !PiecewiseDecay
+    gamma: 0.1
+    milestones: [180000, 240000]
+  - !LinearWarmup
+    start_factor: 0.
+    steps: 1000
+
+OptimizerBuilder:
+  optimizer:
+    momentum: 0.9
+    type: Momentum
+  regularizer:
+    factor: 0.0001
+    type: L2
+
+_READER_: 'solov2_reader.yml'
+TrainReader:
+  sample_transforms:
+  - !DecodeImage
+    to_rgb: true
+  - !Poly2Mask {}
+  - !ResizeImage
+    target_size: [640, 672, 704, 736, 768, 800]
+    max_size: 1333
+    interp: 1
+    use_cv2: true
+    resize_box: true
+  - !RandomFlipImage
+    prob: 0.5
+  - !NormalizeImage
+    is_channel_first: false
+    is_scale: true
+    mean: [0.485,0.456,0.406]
+    std: [0.229, 0.224,0.225]
+  - !Permute
+    to_bgr: false
+    channel_first: true
+  batch_transforms:
+  - !PadBatch
+    pad_to_stride: 32
+  - !Gt2Solov2Target
+    num_grids: [40, 36, 24, 16, 12]
+    scale_ranges: [[1, 96], [48, 192], [96, 384], [192, 768], [384, 2048]]
+    coord_sigma: 0.2
diff --git a/ppdet/modeling/architectures/solov2.py b/ppdet/modeling/architectures/solov2.py
index 756a269c4..76ac0a141 100644
--- a/ppdet/modeling/architectures/solov2.py
+++ b/ppdet/modeling/architectures/solov2.py
@@ -36,24 +36,25 @@ class SOLOv2(object):
         fpn (object): feature pyramid network instance
         bbox_head (object): an `SOLOv2Head` instance
         mask_head (object): an `SOLOv2MaskHead` instance
-        batch_size (int): batch size.
+        train_batch_size (int): training batch size.
     """
 
     __category__ = 'architecture'
     __inject__ = ['backbone', 'fpn', 'bbox_head', 'mask_head']
+    __shared__ = ['train_batch_size']
 
     def __init__(self,
                  backbone,
                  fpn=None,
                  bbox_head='SOLOv2Head',
                  mask_head='SOLOv2MaskHead',
-                 batch_size=1):
+                 train_batch_size=1):
         super(SOLOv2, self).__init__()
         self.backbone = backbone
         self.fpn = fpn
         self.bbox_head = bbox_head
         self.mask_head = mask_head
-        self.batch_size = batch_size
+        self.train_batch_size = train_batch_size
 
     def build(self, feed_vars, mode='train'):
         im = feed_vars['image']
@@ -79,6 +80,8 @@ class SOLOv2(object):
 
         if not mode == 'train':
             self.batch_size = 1
+        else:
+            self.batch_size = self.train_batch_size
 
         mask_feat_pred = self.mask_head.get_output(body_feats, self.batch_size)
 
@@ -103,9 +106,10 @@ class SOLOv2(object):
             cate_preds, kernel_preds = self.bbox_head.get_outputs(
                 body_feats, batch_size=self.batch_size)
 
-            losses = self.bbox_head.get_loss(
-                cate_preds, kernel_preds, mask_feat_pred, ins_labels,
-                cate_labels, grid_orders, fg_num, grid_offset, self.batch_size)
+            losses = self.bbox_head.get_loss(cate_preds, kernel_preds,
+                                             mask_feat_pred, ins_labels,
+                                             cate_labels, grid_orders, fg_num,
+                                             grid_offset, self.train_batch_size)
             total_loss = fluid.layers.sum(list(losses.values()))
             losses.update({'loss': total_loss})
             return losses
-- 
GitLab