diff --git a/configs/mask_rcnn_r50_fpn_1x.yml b/configs/mask_rcnn_r50_fpn_1x.yml
index a1c90e3c02909d426d722d9b1004e9d76e6f8fbc..30b97fde037af92409d283a3b80ce35afa806a79 100644
--- a/configs/mask_rcnn_r50_fpn_1x.yml
+++ b/configs/mask_rcnn_r50_fpn_1x.yml
@@ -130,7 +130,7 @@ LearningRate:
     gamma: 0.1
     milestones: [120000, 160000]
   - !LinearWarmup
-    start_factor: 0.3333
+    start_factor: 0.3333333
     steps: 500
 
 OptimizerBuilder:
diff --git a/configs/yolov3_darknet.yml b/configs/yolov3_darknet.yml
index 7a1215def9fabffef2c2a4de6552636abddf27bd..15e899a9285f048bc32835952d0c604fb8699790 100644
--- a/configs/yolov3_darknet.yml
+++ b/configs/yolov3_darknet.yml
@@ -3,13 +3,13 @@ use_gpu: true
 max_iters: 500000
 log_smooth_window: 20
 save_dir: output
-snapshot_iter: 10000
+snapshot_iter: 50000
 metric: COCO
-pretrain_weights: https://paddlemodels.bj.bcebos.com/yolo/darknet53.pdparams
+pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/DarkNet53_pretrained.tar
 weights: output/yolov3_darknet/model_final
 num_classes: 80
 use_fine_grained_loss: false
-open_debug: False
+load_static_weights: True
 
 YOLOv3:
   anchor: AnchorYOLO
@@ -18,11 +18,15 @@ YOLOv3:
 
 DarkNet:
   depth: 53
+  return_idx: [2, 3, 4]
 
 YOLOv3Head:
   yolo_feat:
     name: YOLOFeat
     feat_in_list: [1024, 768, 384]
+  ignore_thresh: 0.7
+  downsample: 32
+  label_smooth: true
   anchor_per_position: 3
 
 AnchorYOLO:
@@ -30,11 +34,6 @@ AnchorYOLO:
     name: AnchorGeneratorYOLO
     anchors: [10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326]
     anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
-  anchor_target_generator:
-    name: AnchorTargetGeneratorYOLO
-    ignore_thresh: 0.7
-    downsample_ratio: 32
-    label_smooth: true
   anchor_post_process:
     name: BBoxPostProcessYOLO
     # decode -> clip
diff --git a/configs/yolov3_reader.yml b/configs/yolov3_reader.yml
index 2a8463f1e6c2cb598ea4a55c6289f5b04b290d4a..5e11364b0ad831e68513f60ea68ca511e1018a8b 100644
--- a/configs/yolov3_reader.yml
+++ b/configs/yolov3_reader.yml
@@ -27,8 +27,8 @@ TrainReader:
     - !BboxXYXY2XYWH {}
   batch_transforms:
   - !RandomShape
-    sizes: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608]
-    random_inter: True
+     sizes: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608]
+     random_inter: True
   - !NormalizeImage
     mean: [0.485, 0.456, 0.406]
     std: [0.229, 0.224, 0.225]
@@ -50,8 +50,8 @@ TrainReader:
   shuffle: true
   mixup_epoch: 250
   drop_last: true
-  worker_num: 8
-  bufsize: 16
+  worker_num: 4
+  bufsize: 4
   use_process: true
 
 
@@ -81,7 +81,7 @@ EvalReader:
     - !Permute
       to_bgr: false
       channel_first: True
-  batch_size: 8
+  batch_size: 1
   drop_empty: false
   worker_num: 8
   bufsize: 16
diff --git a/ppdet/data/reader.py b/ppdet/data/reader.py
index 7d808b58986a3b0941c007c3bd6621f77ba9c5f5..5cd9f3971136aaa108fda3db39b0615408f951cb 100644
--- a/ppdet/data/reader.py
+++ b/ppdet/data/reader.py
@@ -201,7 +201,7 @@ class Reader(object):
                  use_fine_grained_loss=False,
                  num_classes=80,
                  bufsize=-1,
-                 memsize='3G',
+                 memsize='500M',
                  inputs_def=None,
                  devices_num=1):
         self._dataset = dataset
diff --git a/ppdet/modeling/architecture/yolo.py b/ppdet/modeling/architecture/yolo.py
index c26ba8f5b2bd67d8ae6ffe2bcf549c570e5341d3..f5045f60e245a3acc0f1fe80187af30c0b5e3374 100644
--- a/ppdet/modeling/architecture/yolo.py
+++ b/ppdet/modeling/architecture/yolo.py
@@ -17,38 +17,34 @@ class YOLOv3(BaseArch):
         'yolo_head',
     ]
 
-    def __init__(self, anchor, backbone, yolo_head, *args, **kwargs):
-        super(YOLOv3, self).__init__(*args, **kwargs)
+    def __init__(self, anchor, backbone, yolo_head):
+        super(YOLOv3, self).__init__()
         self.anchor = anchor
         self.backbone = backbone
         self.yolo_head = yolo_head
 
     def model_arch(self, ):
         # Backbone
-        bb_out = self.backbone(self.gbd)
-        self.gbd.update(bb_out)
+        body_feats = self.backbone(self.inputs)
 
         # YOLO Head
-        yolo_head_out = self.yolo_head(self.gbd)
-        self.gbd.update(yolo_head_out)
+        self.yolo_head_out = self.yolo_head(body_feats)
 
         # Anchor
-        anchor_out = self.anchor(self.gbd)
-        self.gbd.update(anchor_out)
-
-        if self.gbd['mode'] == 'infer':
-            bbox_out = self.anchor.post_process(self.gbd)
-            self.gbd.update(bbox_out)
+        self.anchors, self.anchor_masks, self.mask_anchors = self.anchor()
 
     def loss(self, ):
-        yolo_loss = self.yolo_head.loss(self.gbd)
-        out = {'loss': yolo_loss}
-        return out
+        yolo_loss = self.yolo_head.loss(self.inputs, self.yolo_head_out,
+                                        self.anchors, self.anchor_masks,
+                                        self.mask_anchors)
+        return yolo_loss
 
     def infer(self, ):
+        bbox, bbox_num = self.anchor.post_process(
+            self.inputs['im_size'], self.yolo_head_out, self.mask_anchors)
         outs = {
-            "bbox": self.gbd['predicted_bbox'].numpy(),
-            "bbox_nums": self.gbd['predicted_bbox_nums'],
-            'im_id': self.gbd['im_id'].numpy()
+            "bbox": bbox.numpy(),
+            "bbox_num": bbox_num,
+            'im_id': self.inputs['im_id'].numpy()
         }
         return outs
diff --git a/ppdet/modeling/backbone/darknet.py b/ppdet/modeling/backbone/darknet.py
index ed34f0763b55af8fac482b4cbecc8524dc378870..cce2647237890c95dc1d8e9329b55b7cf653f45e 100755
--- a/ppdet/modeling/backbone/darknet.py
+++ b/ppdet/modeling/backbone/darknet.py
@@ -16,7 +16,8 @@ class ConvBNLayer(Layer):
                  stride=1,
                  groups=1,
                  padding=0,
-                 act="leaky"):
+                 act="leaky",
+                 name=None):
         super(ConvBNLayer, self).__init__()
 
         self.conv = Conv2D(
@@ -26,18 +27,18 @@ class ConvBNLayer(Layer):
             stride=stride,
             padding=padding,
             groups=groups,
-            param_attr=ParamAttr(
-                initializer=fluid.initializer.Normal(0., 0.02)),
+            param_attr=ParamAttr(name=name + '.conv.weights'),
             bias_attr=False,
             act=None)
+        bn_name = name + '.bn'
         self.batch_norm = BatchNorm(
             num_channels=ch_out,
             param_attr=ParamAttr(
-                initializer=fluid.initializer.Normal(0., 0.02),
-                regularizer=L2Decay(0.)),
+                name=bn_name + '.scale', regularizer=L2Decay(0.)),
             bias_attr=ParamAttr(
-                initializer=fluid.initializer.Constant(0.0),
-                regularizer=L2Decay(0.)))
+                name=bn_name + '.offset', regularizer=L2Decay(0.)),
+            moving_mean_name=bn_name + '.mean',
+            moving_variance_name=bn_name + '.var')
 
         self.act = act
 
@@ -50,7 +51,13 @@ class ConvBNLayer(Layer):
 
 
 class DownSample(Layer):
-    def __init__(self, ch_in, ch_out, filter_size=3, stride=2, padding=1):
+    def __init__(self,
+                 ch_in,
+                 ch_out,
+                 filter_size=3,
+                 stride=2,
+                 padding=1,
+                 name=None):
 
         super(DownSample, self).__init__()
 
@@ -59,7 +66,8 @@ class DownSample(Layer):
             ch_out=ch_out,
             filter_size=filter_size,
             stride=stride,
-            padding=padding)
+            padding=padding,
+            name=name)
         self.ch_out = ch_out
 
     def forward(self, inputs):
@@ -68,13 +76,23 @@ class DownSample(Layer):
 
 
 class BasicBlock(Layer):
-    def __init__(self, ch_in, ch_out):
+    def __init__(self, ch_in, ch_out, name=None):
         super(BasicBlock, self).__init__()
 
         self.conv1 = ConvBNLayer(
-            ch_in=ch_in, ch_out=ch_out, filter_size=1, stride=1, padding=0)
+            ch_in=ch_in,
+            ch_out=ch_out,
+            filter_size=1,
+            stride=1,
+            padding=0,
+            name=name + '.0')
         self.conv2 = ConvBNLayer(
-            ch_in=ch_out, ch_out=ch_out * 2, filter_size=3, stride=1, padding=1)
+            ch_in=ch_out,
+            ch_out=ch_out * 2,
+            filter_size=3,
+            stride=1,
+            padding=1,
+            name=name + '.1')
 
     def forward(self, inputs):
         conv1 = self.conv1(inputs)
@@ -84,14 +102,16 @@ class BasicBlock(Layer):
 
 
 class Blocks(Layer):
-    def __init__(self, ch_in, ch_out, count):
+    def __init__(self, ch_in, ch_out, count, name=None):
         super(Blocks, self).__init__()
 
-        self.basicblock0 = BasicBlock(ch_in, ch_out)
+        self.basicblock0 = BasicBlock(ch_in, ch_out, name=name + '.0')
         self.res_out_list = []
         for i in range(1, count):
-            res_out = self.add_sublayer("basic_block_%d" % (i),
-                                        BasicBlock(ch_out * 2, ch_out))
+            block_name = '{}.{}'.format(name, i)
+            res_out = self.add_sublayer(
+                block_name, BasicBlock(
+                    ch_out * 2, ch_out, name=block_name))
             self.res_out_list.append(res_out)
         self.ch_out = ch_out
 
@@ -108,31 +128,46 @@ DarkNet_cfg = {53: ([1, 2, 8, 8, 4])}
 @register
 @serializable
 class DarkNet(Layer):
-    def __init__(self, depth=53, mode='train'):
+    def __init__(self,
+                 depth=53,
+                 freeze_at=-1,
+                 return_idx=[2, 3, 4],
+                 num_stages=5):
         super(DarkNet, self).__init__()
         self.depth = depth
-        self.mode = mode
-        self.stages = DarkNet_cfg[self.depth][0:5]
+        self.freeze_at = freeze_at
+        self.return_idx = return_idx
+        self.num_stages = num_stages
+        self.stages = DarkNet_cfg[self.depth][0:num_stages]
 
         self.conv0 = ConvBNLayer(
-            ch_in=3, ch_out=32, filter_size=3, stride=1, padding=1)
+            ch_in=3,
+            ch_out=32,
+            filter_size=3,
+            stride=1,
+            padding=1,
+            name='yolo_input')
 
-        self.downsample0 = DownSample(ch_in=32, ch_out=32 * 2)
+        self.downsample0 = DownSample(
+            ch_in=32, ch_out=32 * 2, name='yolo_input.downsample')
 
-        self.darknet53_conv_block_list = []
+        self.darknet_conv_block_list = []
         self.downsample_list = []
         ch_in = [64, 128, 256, 512, 1024]
         for i, stage in enumerate(self.stages):
-            conv_block = self.add_sublayer("stage_%d" % (i),
-                                           Blocks(
-                                               int(ch_in[i]), 32 * (2**i),
-                                               stage))
-            self.darknet53_conv_block_list.append(conv_block)
-        for i in range(len(self.stages) - 1):
+            name = 'stage.{}'.format(i)
+            conv_block = self.add_sublayer(
+                name, Blocks(
+                    int(ch_in[i]), 32 * (2**i), stage, name=name))
+            self.darknet_conv_block_list.append(conv_block)
+        for i in range(num_stages - 1):
+            down_name = 'stage.{}.downsample'.format(i)
             downsample = self.add_sublayer(
-                "stage_%d_downsample" % i,
+                down_name,
                 DownSample(
-                    ch_in=32 * (2**(i + 1)), ch_out=32 * (2**(i + 2))))
+                    ch_in=32 * (2**(i + 1)),
+                    ch_out=32 * (2**(i + 2)),
+                    name=down_name))
             self.downsample_list.append(downsample)
 
     def forward(self, inputs):
@@ -141,10 +176,12 @@ class DarkNet(Layer):
         out = self.conv0(x)
         out = self.downsample0(out)
         blocks = []
-        for i, conv_block_i in enumerate(self.darknet53_conv_block_list):
+        for i, conv_block_i in enumerate(self.darknet_conv_block_list):
             out = conv_block_i(out)
-            blocks.append(out)
-            if i < len(self.stages) - 1:
+            if i == self.freeze_at:
+                out.stop_gradient = True
+            if i in self.return_idx:
+                blocks.append(out)
+            if i < self.num_stages - 1:
                 out = self.downsample_list[i](out)
-        outs = {'darknet_outs': blocks[-1:-4:-1]}
-        return outs
+        return blocks
diff --git a/ppdet/modeling/bbox.py b/ppdet/modeling/bbox.py
index 235f4dc55eb54bf42297e8f0768307c98419eed6..fefc3b51efc9da7e9cb4c2bf6907be95e9106140 100644
--- a/ppdet/modeling/bbox.py
+++ b/ppdet/modeling/bbox.py
@@ -79,27 +79,25 @@ class BBoxPostProcessYOLO(object):
         self.decode = decode
         self.clip = clip
 
-    def __call__(self, inputs):
+    def __call__(self, im_size, yolo_head_out, mask_anchors):
         # TODO: split yolo_box into 2 steps
         # decode
         # clip
         boxes_list = []
         scores_list = []
-        for i, out in enumerate(inputs['yolo_outs']):
-            boxes, scores = self.yolo_box(out, inputs['im_size'],
-                                          inputs['mask_anchors'][i], i,
-                                          "yolo_box_" + str(i))
+        for i, head_out in enumerate(yolo_head_out):
+            boxes, scores = self.yolo_box(head_out, im_size, mask_anchors[i],
+                                          self.num_classes, i)
 
             boxes_list.append(boxes)
             scores_list.append(fluid.layers.transpose(scores, perm=[0, 2, 1]))
         yolo_boxes = fluid.layers.concat(boxes_list, axis=1)
         yolo_scores = fluid.layers.concat(scores_list, axis=2)
-        nmsed_bbox = self.nms(bboxes=yolo_boxes, scores=yolo_scores)
+        bbox = self.nms(bboxes=yolo_boxes, scores=yolo_scores)
         # TODO: parse the lod of nmsed_bbox
         # default batch size is 1
-        bbox_nums = np.array([0, int(nmsed_bbox.shape[0])], dtype=np.int32)
-        outs = {"predicted_bbox_nums": bbox_nums, "predicted_bbox": nmsed_bbox}
-        return outs
+        bbox_num = np.array([int(bbox.shape[0])], dtype=np.int32)
+        return bbox, bbox_num
 
 
 @register
@@ -168,32 +166,18 @@ class AnchorRPN(object):
 
 @register
 class AnchorYOLO(object):
-    __inject__ = [
-        'anchor_generator', 'anchor_target_generator', 'anchor_post_process'
-    ]
+    __inject__ = ['anchor_generator', 'anchor_post_process']
 
-    def __init__(self, anchor_generator, anchor_target_generator,
-                 anchor_post_process):
+    def __init__(self, anchor_generator, anchor_post_process):
         super(AnchorYOLO, self).__init__()
         self.anchor_generator = anchor_generator
-        self.anchor_target_generator = anchor_target_generator
         self.anchor_post_process = anchor_post_process
 
-    def __call__(self, inputs):
-        outs = self.generate_anchors(inputs)
-        return outs
-
-    def generate_anchors(self, inputs):
-        outs = self.anchor_generator(inputs['yolo_outs'])
-        outs['anchor_module'] = self
-        return outs
-
-    def generate_anchors_target(self, inputs):
-        outs = self.anchor_target_generator()
-        return outs
+    def __call__(self):
+        return self.anchor_generator()
 
-    def post_process(self, inputs):
-        return self.anchor_post_process(inputs)
+    def post_process(self, im_size, yolo_head_out, mask_anchors):
+        return self.anchor_post_process(im_size, yolo_head_out, mask_anchors)
 
 
 @register
diff --git a/ppdet/modeling/head/mask_head.py b/ppdet/modeling/head/mask_head.py
index 3ab92daa4096761987db9b2b52965eb3da72fa67..47f202a1b99d9da73962ed13ebbe157eccfd738e 100644
--- a/ppdet/modeling/head/mask_head.py
+++ b/ppdet/modeling/head/mask_head.py
@@ -37,7 +37,7 @@ class MaskFeat(Layer):
                 mask_conv.add_sublayer(
                     conv_name,
                     Conv2D(
-                        num_channels=feat_in if j == 1 else feat_out,
+                        num_channels=feat_in if j == 0 else feat_out,
                         num_filters=feat_out,
                         filter_size=3,
                         act='relu',
diff --git a/ppdet/modeling/head/yolo_head.py b/ppdet/modeling/head/yolo_head.py
index 56b1bafab6011326ab3601a31fd54f7bc5fdfcd6..cae1f4024e889464d9401d5a0b1d5ca57e1f1b56 100644
--- a/ppdet/modeling/head/yolo_head.py
+++ b/ppdet/modeling/head/yolo_head.py
@@ -1,150 +1,135 @@
 import paddle.fluid as fluid
+import paddle
 from paddle.fluid.dygraph import Layer
 from paddle.fluid.param_attr import ParamAttr
 from paddle.fluid.initializer import Normal
 from paddle.fluid.regularizer import L2Decay
 from paddle.fluid.dygraph.nn import Conv2D, BatchNorm
+from paddle.fluid.dygraph import Sequential
 from ppdet.core.workspace import register
 from ..backbone.darknet import ConvBNLayer
 
 
 class YoloDetBlock(Layer):
-    def __init__(self, ch_in, channel):
+    def __init__(self, ch_in, channel, name):
         super(YoloDetBlock, self).__init__()
-
+        self.ch_in = ch_in
+        self.channel = channel
         assert channel % 2 == 0, \
             "channel {} cannot be divided by 2".format(channel)
-
-        self.conv0 = ConvBNLayer(
-            ch_in=ch_in, ch_out=channel, filter_size=1, stride=1, padding=0)
-
-        self.conv1 = ConvBNLayer(
-            ch_in=channel,
-            ch_out=channel * 2,
-            filter_size=3,
-            stride=1,
-            padding=1)
-
-        self.conv2 = ConvBNLayer(
-            ch_in=channel * 2,
-            ch_out=channel,
-            filter_size=1,
-            stride=1,
-            padding=0)
-
-        self.conv3 = ConvBNLayer(
-            ch_in=channel,
-            ch_out=channel * 2,
-            filter_size=3,
-            stride=1,
-            padding=1)
-
-        self.route = ConvBNLayer(
-            ch_in=channel * 2,
-            ch_out=channel,
-            filter_size=1,
-            stride=1,
-            padding=0)
+        conv_def = [
+            ['conv0', ch_in, channel, 1, '.0.0'],
+            ['conv1', channel, channel * 2, 3, '.0.1'],
+            ['conv2', channel * 2, channel, 1, '.1.0'],
+            ['conv3', channel, channel * 2, 3, '.1.1'],
+            ['route', channel * 2, channel, 1, '.2'],
+            #['tip', channel, channel * 2, 3],
+        ]
+
+        self.conv_module = Sequential()
+        for idx, (conv_name, ch_in, ch_out, filter_size,
+                  post_name) in enumerate(conv_def):
+            self.conv_module.add_sublayer(
+                conv_name,
+                ConvBNLayer(
+                    ch_in=ch_in,
+                    ch_out=ch_out,
+                    filter_size=filter_size,
+                    padding=(filter_size - 1) // 2,
+                    name=name + post_name))
 
         self.tip = ConvBNLayer(
             ch_in=channel,
             ch_out=channel * 2,
             filter_size=3,
-            stride=1,
-            padding=1)
+            padding=1,
+            name=name + '.tip')
 
     def forward(self, inputs):
-        out = self.conv0(inputs)
-        out = self.conv1(out)
-        out = self.conv2(out)
-        out = self.conv3(out)
-        route = self.route(out)
+        route = self.conv_module(inputs)
         tip = self.tip(route)
         return route, tip
 
 
-class Upsample(Layer):
-    def __init__(self, scale=2):
-        super(Upsample, self).__init__()
-        self.scale = scale
-
-    def forward(self, inputs):
-        # get dynamic upsample output shape
-        shape_nchw = fluid.layers.shape(inputs)
-        shape_hw = fluid.layers.slice(
-            shape_nchw, axes=[0], starts=[2], ends=[4])
-        shape_hw.stop_gradient = True
-        in_shape = fluid.layers.cast(shape_hw, dtype='int32')
-        out_shape = in_shape * self.scale
-        out_shape.stop_gradient = True
-
-        # reisze by actual_shape
-        out = fluid.layers.resize_nearest(
-            input=inputs, scale=self.scale, actual_shape=out_shape)
-        return out
-
-
 @register
 class YOLOFeat(Layer):
-    def __init__(self, feat_in_list=[1024, 768, 384]):
+    __shared__ = ['num_levels']
+
+    def __init__(self, feat_in_list=[1024, 768, 384], num_levels=3):
         super(YOLOFeat, self).__init__()
         self.feat_in_list = feat_in_list
         self.yolo_blocks = []
         self.route_blocks = []
-        for i in range(3):
+        self.num_levels = num_levels
+        for i in range(self.num_levels):
+            name = 'yolo_block.{}'.format(i)
             yolo_block = self.add_sublayer(
-                "yolo_det_block_%d" % (i),
+                name,
                 YoloDetBlock(
-                    feat_in_list[i], channel=512 // (2**i)))
+                    feat_in_list[i], channel=512 // (2**i), name=name))
             self.yolo_blocks.append(yolo_block)
 
-            if i < 2:
+            if i < self.num_levels - 1:
+                name = 'yolo_transition.{}'.format(i)
                 route = self.add_sublayer(
-                    "route_%d" % i,
+                    name,
                     ConvBNLayer(
                         ch_in=512 // (2**i),
                         ch_out=256 // (2**i),
                         filter_size=1,
                         stride=1,
-                        padding=0))
+                        padding=0,
+                        name=name))
                 self.route_blocks.append(route)
-        self.upsample = Upsample()
 
-    def forward(self, inputs):
+    def forward(self, body_feats):
+        assert len(body_feats) == self.num_levels
+        body_feats = body_feats[::-1]
         yolo_feats = []
-        for i, block in enumerate(inputs['darknet_outs']):
+        for i, block in enumerate(body_feats):
             if i > 0:
                 block = fluid.layers.concat(input=[route, block], axis=1)
-
             route, tip = self.yolo_blocks[i](block)
             yolo_feats.append(tip)
 
-            if i < 2:
+            if i < self.num_levels - 1:
                 route = self.route_blocks[i](route)
-                route = self.upsample(route)
+                route = fluid.layers.resize_nearest(route, scale=2.)
 
-        outs = {'yolo_feat': yolo_feats}
-        return outs
+        return yolo_feats
 
 
 @register
 class YOLOv3Head(Layer):
-    __shared__ = ['num_classes']
+    __shared__ = ['num_classes', 'num_levels', 'use_fine_grained_loss']
     __inject__ = ['yolo_feat']
 
-    def __init__(self, yolo_feat, num_classes=80, anchor_per_position=3):
+    def __init__(self,
+                 yolo_feat,
+                 num_classes=80,
+                 anchor_per_position=3,
+                 num_levels=3,
+                 use_fine_grained_loss=False,
+                 ignore_thresh=0.7,
+                 downsample=32,
+                 label_smooth=True):
         super(YOLOv3Head, self).__init__()
         self.num_classes = num_classes
         self.anchor_per_position = anchor_per_position
         self.yolo_feat = yolo_feat
-
-        self.yolo_outs = []
-        for i in range(3):
+        self.num_levels = num_levels
+        self.use_fine_grained_loss = use_fine_grained_loss
+        self.ignore_thresh = ignore_thresh
+        self.downsample = downsample
+        self.label_smooth = label_smooth
+        self.yolo_out_list = []
+        for i in range(num_levels):
             # TODO: optim here
             #num_filters = len(cfg.anchor_masks[i]) * (self.num_classes + 5)
             num_filters = self.anchor_per_position * (self.num_classes + 5)
+            name = 'yolo_output.{}'.format(i)
             yolo_out = self.add_sublayer(
-                "yolo_out_%d" % (i),
+                name,
                 Conv2D(
                     num_channels=1024 // (2**i),
                     num_filters=num_filters,
@@ -152,44 +137,38 @@ class YOLOv3Head(Layer):
                     stride=1,
                     padding=0,
                     act=None,
-                    param_attr=ParamAttr(
-                        initializer=fluid.initializer.Normal(0., 0.02)),
+                    param_attr=ParamAttr(name=name + '.conv.weights'),
                     bias_attr=ParamAttr(
-                        initializer=fluid.initializer.Constant(0.0),
-                        regularizer=L2Decay(0.))))
-            self.yolo_outs.append(yolo_out)
+                        name=name + '.conv.bias', regularizer=L2Decay(0.))))
+            self.yolo_out_list.append(yolo_out)
+
+    def forward(self, body_feats):
+        assert len(body_feats) == self.num_levels
+        yolo_feats = self.yolo_feat(body_feats)
+        yolo_head_out = []
+        for i, feat in enumerate(yolo_feats):
+            yolo_out = self.yolo_out_list[i](feat)
+            yolo_head_out.append(yolo_out)
+        return yolo_head_out
+
+    def loss(self, inputs, head_out, anchors, anchor_masks, mask_anchors):
+        if self.use_fine_grained_loss:
+            raise NotImplementedError
 
-    def forward(self, inputs):
-        outs = self.yolo_feat(inputs)
-        x = outs['yolo_feat']
-        yolo_out_list = []
-        for i, yolo_f in enumerate(x):
-            yolo_out = self.yolo_outs[i](yolo_f)
-            yolo_out_list.append(yolo_out)
-        outs.update({"yolo_outs": yolo_out_list})
-        return outs
-
-    def loss(self, inputs):
-        if callable(inputs['anchor_module']):
-            yolo_targets = inputs['anchor_module'].generate_anchors_target(
-                inputs)
         yolo_losses = []
-        for i, out in enumerate(inputs['yolo_outs']):
-            # TODO: split yolov3_loss into small ops
-            # 1. compute target 2. loss
+        for i, out in enumerate(head_out):
             loss = fluid.layers.yolov3_loss(
                 x=out,
                 gt_box=inputs['gt_bbox'],
                 gt_label=inputs['gt_class'],
                 gt_score=inputs['gt_score'],
-                anchors=inputs['anchors'],
-                anchor_mask=inputs['anchor_masks'][i],
+                anchors=anchors,
+                anchor_mask=anchor_masks[i],
                 class_num=self.num_classes,
-                ignore_thresh=yolo_targets['ignore_thresh'],
-                downsample_ratio=yolo_targets['downsample_ratio'] // 2**i,
-                use_label_smooth=yolo_targets['label_smooth'],
+                ignore_thresh=self.ignore_thresh,
+                downsample_ratio=self.downsample // 2**i,
+                use_label_smooth=self.label_smooth,
                 name='yolo_loss_' + str(i))
             loss = fluid.layers.reduce_mean(loss)
             yolo_losses.append(loss)
-        yolo_loss = sum(yolo_losses)
-        return yolo_loss
+        return {'loss': sum(yolo_losses)}
diff --git a/ppdet/modeling/ops.py b/ppdet/modeling/ops.py
index 5d7a112939a9ddbde947f6aa78507b34edcc732a..7f2edbac1d894828847536e4ffd2ee2232c161ae 100644
--- a/ppdet/modeling/ops.py
+++ b/ppdet/modeling/ops.py
@@ -99,20 +99,16 @@ class AnchorGeneratorYOLO(object):
         self.anchors = anchors
         self.anchor_masks = anchor_masks
 
-    def __call__(self, yolo_outs):
+    def __call__(self):
+        anchor_num = len(self.anchors)
         mask_anchors = []
-        for i, _ in enumerate(yolo_outs):
+        for i in range(len(self.anchor_masks)):
             mask_anchor = []
             for m in self.anchor_masks[i]:
-                mask_anchor.append(self.anchors[2 * m])
-                mask_anchor.append(self.anchors[2 * m + 1])
+                assert m < anchor_num, "anchor mask index overflow"
+                mask_anchor.extend(self.anchors[2 * m:2 * m + 2])
             mask_anchors.append(mask_anchor)
-        outs = {
-            "anchors": self.anchors,
-            "anchor_masks": self.anchor_masks,
-            "mask_anchors": mask_anchors
-        }
-        return outs
+        return self.anchors, self.anchor_masks, mask_anchors
 
 
 @register
@@ -305,6 +301,7 @@ class RoIExtractor(object):
             self.canconical_level,
             self.canonical_size,
             rois_num=rois_num)
+
         rois_feat_list = []
         for lvl in range(self.start_level, self.end_level + 1):
             roi_feat = fluid.layers.roi_align(
@@ -381,24 +378,19 @@ class MultiClassNMS(object):
 @register
 @serializable
 class YOLOBox(object):
-    __shared__ = ['num_classes']
-
     def __init__(
             self,
-            num_classes=80,
             conf_thresh=0.005,
             downsample_ratio=32,
             clip_bbox=True, ):
-        self.num_classes = num_classes
         self.conf_thresh = conf_thresh
         self.downsample_ratio = downsample_ratio
         self.clip_bbox = clip_bbox
 
-    def __call__(self, x, img_size, anchors, stage=0, name=None):
-
-        outs = fluid.layers.yolo_box(x, img_size, anchors, self.num_classes,
+    def __call__(self, x, img_size, anchors, num_classes, stage=0):
+        outs = fluid.layers.yolo_box(x, img_size, anchors, num_classes,
                                      self.conf_thresh, self.downsample_ratio //
-                                     2**stage, self.clip_bbox, name)
+                                     2**stage, self.clip_bbox)
         return outs
 
 
diff --git a/ppdet/optimizer.py b/ppdet/optimizer.py
index 2016cda13384385aeec88b667d7543090d66f24a..b2c73c88d1cdaa0d6c30032d9926a0e38142e921 100644
--- a/ppdet/optimizer.py
+++ b/ppdet/optimizer.py
@@ -82,7 +82,7 @@ class LinearWarmup(object):
     def __call__(self, base_lr):
         boundary = []
         value = []
-        for i in range(self.steps):
+        for i in range(self.steps + 1):
             alpha = i / self.steps
             factor = self.start_factor * (1 - alpha) + alpha
             lr = base_lr * factor
diff --git a/ppdet/py_op/bbox.py b/ppdet/py_op/bbox.py
index dc34a77cd2b3f09a76fb0ca1476458855ebcc301..33f0c8837658d73609a6e9d6be37de2eae399b51 100755
--- a/ppdet/py_op/bbox.py
+++ b/ppdet/py_op/bbox.py
@@ -190,7 +190,7 @@ def nms_with_decode(bboxes,
         rois_n = bboxes_v[start:end, :]  # box 
         rois_n = rois_n / im_info[i][2]  # scale 
         rois_n = delta2bbox(bbox_deltas_n, rois_n, variance_v)
-        rois_n = clip_bbox(rois_n, im_info[i][:2] / im_info[i][2])
+        rois_n = clip_bbox(rois_n, np.round(im_info[i][:2] / im_info[i][2]))
         cls_boxes = [[] for _ in range(class_nums)]
         scores_n = bbox_probs_v[start:end, :]
         for j in range(1, class_nums):
diff --git a/ppdet/py_op/post_process.py b/ppdet/py_op/post_process.py
index c6bf354e47453fade8c9fc88dea42f92df90b478..2049bffd0df80e45d61eba35c3ce77395ac8ad42 100755
--- a/ppdet/py_op/post_process.py
+++ b/ppdet/py_op/post_process.py
@@ -136,7 +136,8 @@ def get_det_res(bboxes, bbox_nums, image_id, num_id_to_cat_id_map):
     k = 0
     for i in range(len(bbox_nums)):
         image_id = int(image_id[i][0])
-
+        if bboxes.shape == (1, 1):
+            continue
         det_nums = bbox_nums[i]
         for j in range(det_nums):
             dt = bboxes[k]
diff --git a/ppdet/utils/checkpoint.py b/ppdet/utils/checkpoint.py
index c2c41947f4cf1c76eeea4a72f5e9178e129058e9..419f5d6112db6390cf94106df194d1b532b1fd56 100644
--- a/ppdet/utils/checkpoint.py
+++ b/ppdet/utils/checkpoint.py
@@ -3,10 +3,12 @@ from __future__ import division
 from __future__ import print_function
 from __future__ import unicode_literals
 
+import errno
 import os
 import time
 import re
 import numpy as np
+import paddle
 import paddle.fluid as fluid
 from .download import get_weights_path
 
@@ -88,9 +90,10 @@ def load_dygraph_ckpt(model,
     return model
 
 
-def save_dygraph_ckpt(model, optimizer, save_dir):
+def save_dygraph_ckpt(model, optimizer, save_dir, save_name):
     if not os.path.exists(save_dir):
         os.makedirs(save_dir)
-    fluid.dygraph.save_dygraph(model.state_dict(), save_dir)
-    fluid.dygraph.save_dygraph(optimizer.state_dict(), save_dir)
+    save_path = os.path.join(save_dir, save_name)
+    fluid.dygraph.save_dygraph(model.state_dict(), save_path)
+    fluid.dygraph.save_dygraph(optimizer.state_dict(), save_path)
     print("Save checkpoint:", save_dir)
diff --git a/tools/eval.py b/tools/eval.py
index d437cd7cb5b3d3e466f66e3846a9b171ed3d1181..b39573d01873d921c798c98f9ab70b99de4f2483 100755
--- a/tools/eval.py
+++ b/tools/eval.py
@@ -20,6 +20,10 @@ from ppdet.utils.cli import ArgsParser
 from ppdet.utils.eval_utils import coco_eval_results
 from ppdet.data.reader import create_reader
 from ppdet.utils.checkpoint import load_dygraph_ckpt, save_dygraph_ckpt
+import logging
+FORMAT = '%(asctime)s-%(levelname)s: %(message)s'
+logging.basicConfig(level=logging.INFO, format=FORMAT)
+logger = logging.getLogger(__name__)
 
 
 def parse_args():
@@ -58,22 +62,26 @@ def run(FLAGS, cfg):
 
     # Run Eval
     outs_res = []
+    start_time = time.time()
+    sample_num = 0
     for iter_id, data in enumerate(eval_reader()):
-        start_time = time.time()
-
         # forward 
         model.eval()
         outs = model(data, cfg['EvalReader']['inputs_def']['fields'], 'infer')
         outs_res.append(outs)
 
         # log 
-        cost_time = time.time() - start_time
-        print("Eval iter: {}, time: {}".format(iter_id, cost_time))
+        sample_num += len(data)
+        if iter_id % 100 == 0:
+            logger.info("Eval iter: {}".format(iter_id))
 
+    cost_time = time.time() - start_time
+    logger.info('Total sample number: {}, averge FPS: {}'.format(
+        sample_num, sample_num / cost_time))
     # Metric 
     coco_eval_results(
         outs_res,
-        include_mask=True if 'MaskHead' in cfg else False,
+        include_mask=True if getattr(cfg, 'MaskHead', None) else False,
         dataset=cfg['EvalReader']['dataset'])
 
 
diff --git a/tools/train.py b/tools/train.py
index 27f13323be3ab7d4f8b82a6d01f71a2a0fcc9ade..76a860909dd2e1c6a593e441e34d4103082cf910 100755
--- a/tools/train.py
+++ b/tools/train.py
@@ -113,7 +113,7 @@ def run(FLAGS, cfg):
         optimizer,
         cfg.pretrain_weights,
         ckpt_type=FLAGS.ckpt_type,
-        load_static_weights=cfg.load_static_weights)
+        load_static_weights=cfg.get('load_static_weights', False))
 
     # Parallel Model 
     if ParallelEnv().nranks > 1:
@@ -177,8 +177,8 @@ def run(FLAGS, cfg):
                 cfg_name = os.path.basename(FLAGS.config).split('.')[0]
                 save_name = str(
                     iter_id) if iter_id != cfg.max_iters - 1 else "model_final"
-                save_dir = os.path.join(cfg.save_dir, cfg_name, save_name)
-                save_dygraph_ckpt(model, optimizer, save_dir)
+                save_dir = os.path.join(cfg.save_dir, cfg_name)
+                save_dygraph_ckpt(model, optimizer, save_dir, save_name)
 
 
 def main():