first about DataLoader and Trainer

deployment/tensorrt-yolov6/README.md

@@ -16,8 +16,8 @@
 1. 下载 yolort 源码:
 
    ```sh
-   git clone https://github.com/zhiqwang/yolort.git
-   cd yolort/deployment/tensorrt-yolov6
+   git clone https://github.com/zhiqwang/yolov5-rt-stack.git
+   cd yolov5-rt-stack/deployment/tensorrt-yolov6
    ```
 
 1. 导出 ONNX 的时候请务必包含 EfficientNMS 算子, 稍后我们会加入一个更清晰的流程. ~从 YOLOv6 官方地址下载 ONNX 模型，如 [yolov6n.onnx](https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6n.onnx)~.

deployment/tensorrt/README.md

@@ -11,7 +11,7 @@ The TensorRT inference example of `yolort`.
 
 ## Usage
 
-Here we will mainly discuss how to use the C++ interface, we recommend that you check out our [tutorial](https://zhiqwang.com/yolort/notebooks/onnx-graphsurgeon-inference-tensorrt.html) first.
+Here we will mainly discuss how to use the C++ interface, we recommend that you check out our [tutorial](https://zhiqwang.com/yolov5-rt-stack/notebooks/onnx-graphsurgeon-inference-tensorrt.html) first.
 
 1. Export your custom model to TensorRT format
 

docs/source/conf.py

@@ -80,7 +80,7 @@ def setup(app):
 
 nbsphinx_epilog = """
 View this document as a notebook:
-https://github.com/zhiqwang/yolort/blob/main/{{ env.doc2path(env.docname, base=None) }}
+https://github.com/zhiqwang/yolov5-rt-stack/blob/main/{{ env.doc2path(env.docname, base=None) }}
 
 ----
 """
@@ -106,12 +106,12 @@ def setup(app):
     # "google_analytics_account": "UA-XXXXX",
     # Specify a base_url used to generate sitemap.xml. If not
     # specified, then no sitemap will be built.
-    "base_url": "https://zhiqwang.com/yolort",
+    "base_url": "https://zhiqwang.com/yolov5-rt-stack",
     # Set the color and the accent color
     "color_primary": "blue",
     "color_accent": "light-blue",
     # Set the repo location to get a badge with stats
-    "repo_url": "https://github.com/zhiqwang/yolort/",
+    "repo_url": "https://github.com/zhiqwang/yolov5-rt-stack/",
     "repo_name": "yolort",
     # Visible levels of the global TOC; -1 means unlimited
     "globaltoc_depth": 3,

docs/source/index.rst

@@ -50,8 +50,8 @@ Read a source of image(s) and detect its objects:
 And we support loading the trained weights from YOLOv5. Please see our documents on what
 we `share`_ and how we `differ`_ from yolov5 for more details.
 
-.. _share: https://zhiqwang.com/yolort/notebooks/how-to-align-with-ultralytics-yolov5.html
-.. _differ: https://zhiqwang.com/yolort/notebooks/comparison-between-yolort-vs-yolov5.html
+.. _share: https://zhiqwang.com/yolov5-rt-stack/notebooks/how-to-align-with-ultralytics-yolov5.html
+.. _differ: https://zhiqwang.com/yolov5-rt-stack/notebooks/comparison-between-yolort-vs-yolov5.html
 
 .. code:: python
 

docs/source/installation.rst

@@ -10,7 +10,7 @@ install PyTorch 1.8.0+ and torchvision 0.9.0+.
 
 To install yolort, you may either use the repository
 
-   https://github.com/zhiqwang/yolort/
+   https://github.com/zhiqwang/yolov5-rt-stack/
 
 and ``setup.py`` from there or use
 
@@ -22,7 +22,7 @@ to get the latest release and
 
 ::
 
-   pip install 'git+https://github.com/zhiqwang/yolort.git'
+   pip install 'git+https://github.com/zhiqwang/yolov5-rt-stack.git'
 
 to install the development version.
 

exps/default/__init__.py

@@ -0,0 +1,3 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.

exps/default/yolov3.py

@@ -0,0 +1,35 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+import torch.nn as nn
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 1.0
+        self.width = 1.0
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]
+
+    def get_model(self, sublinear=False):
+        def init_yolo(M):
+            for m in M.modules():
+                if isinstance(m, nn.BatchNorm2d):
+                    m.eps = 1e-3
+                    m.momentum = 0.03
+
+        if "model" not in self.__dict__:
+            from yolort.models import YOLOFPN, YOLOX, YOLOXHead
+
+            backbone = YOLOFPN()
+            head = YOLOXHead(self.num_classes, self.width, in_channels=[128, 256, 512], act="lrelu")
+            self.model = YOLOX(backbone, head)
+        self.model.apply(init_yolo)
+        self.model.head.initialize_biases(1e-2)
+
+        return self.model

exps/default/yolox_l.py

@@ -0,0 +1,15 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 1.0
+        self.width = 1.0
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]

exps/default/yolox_m.py

@@ -0,0 +1,15 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 0.67
+        self.width = 0.75
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]

exps/default/yolox_nano.py

@@ -0,0 +1,51 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+import torch.nn as nn
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 0.33
+        self.width = 0.25
+        self.input_size = (416, 416)
+        self.random_size = (10, 20)
+        self.mosaic_scale = (0.5, 1.5)
+        self.test_size = (416, 416)
+        self.mosaic_prob = 0.5
+        self.enable_mixup = False
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]
+
+    def get_model(self, sublinear=False):
+        def init_yolo(M):
+            for m in M.modules():
+                if isinstance(m, nn.BatchNorm2d):
+                    m.eps = 1e-3
+                    m.momentum = 0.03
+
+        if "model" not in self.__dict__:
+            from yolort.models import YOLOPAFPN, YOLOX, YOLOXHead
+
+            in_channels = [256, 512, 1024]
+            # NANO model use depthwise = True, which is main difference.
+            backbone = YOLOPAFPN(
+                self.depth,
+                self.width,
+                in_channels=in_channels,
+                act=self.act,
+                depthwise=True,
+            )
+            head = YOLOXHead(
+                self.num_classes, self.width, in_channels=in_channels, act=self.act, depthwise=True
+            )
+            self.model = YOLOX(backbone, head)
+
+        self.model.apply(init_yolo)
+        self.model.head.initialize_biases(1e-2)
+        return self.model

exps/default/yolox_s.py

@@ -0,0 +1,15 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 0.33
+        self.width = 0.50
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]

exps/default/yolox_tiny.py

@@ -0,0 +1,20 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 0.33
+        self.width = 0.375
+        self.input_size = (416, 416)
+        self.mosaic_scale = (0.5, 1.5)
+        self.random_size = (10, 20)
+        self.test_size = (416, 416)
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]
+        self.enable_mixup = False

exps/default/yolox_x.py

@@ -0,0 +1,15 @@
+#!/usr/bin/env python3
+# -*- coding:utf-8 -*-
+# Copyright (c) Megvii, Inc. and its affiliates.
+
+import os
+
+from yolort.exp import Exp as MyExp
+
+
+class Exp(MyExp):
+    def __init__(self):
+        super(Exp, self).__init__()
+        self.depth = 1.33
+        self.width = 1.25
+        self.exp_name = os.path.split(os.path.realpath(__file__))[1].split(".")[0]

notebooks/comparison-between-yolort-vs-yolov5.ipynb

@@ -11,7 +11,7 @@
     "\n",
     "For pre-processing, YOLOv5 uses the [letterboxing (padding)](https://en.wikipedia.org/wiki/Letterboxing_(filming)) resizing that maintains the aspect ratio. The error arises from the `interpolate` operator used in resizing. YOLOv5 uses the [cv2.resize](https://docs.opencv.org/4.x/da/d54/group__imgproc__transform.html#ga47a974309e9102f5f08231edc7e7529d) operator on the input `uint8 [0-255]` images, the operators in OpenCV are not traceable or scriptable, so we use [torch.nn.functional.interpolate](https://pytorch.org/docs/stable/generated/torch.nn.functional.interpolate.html#torch-nn-functional-interpolate) in yolort. Fortunately, the `interpolation` operator of PyTorch is [aligned](https://github.com/pytorch/vision/issues/2950#issuecomment-723038041) with that of OpenCV, however PyTorch's `interpolate` [only supports](https://github.com/pytorch/pytorch/issues/5580) the float data type now, we can only operate with images cast to float types, and therefore there will introduce some errors.\n",
     "\n",
-    "YOLOv5 provides a very powerful function to do the post-processing, of which we implement only a non-agnostic version, but the accuracy here should be able to be consistent with the original version. See our [doc](https://zhiqwang.com/yolort/notebooks/how-to-align-with-ultralytics-yolov5.html) for more details.\n",
+    "YOLOv5 provides a very powerful function to do the post-processing, of which we implement only a non-agnostic version, but the accuracy here should be able to be consistent with the original version. See our [doc](https://zhiqwang.com/yolov5-rt-stack/notebooks/how-to-align-with-ultralytics-yolov5.html) for more details.\n",
     "\n",
     "## Prepare environment, image and model weights to test"
    ]

notebooks/export-onnx-inference-onnxruntime.ipynb

@@ -342,15 +342,15 @@
      "text": [
       "/opt/conda/lib/python3.8/site-packages/torch/nn/functional.py:3701: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
       "  (torch.floor((input.size(i + 2).float() * torch.tensor(scale_factors[i], dtype=torch.float32)).float()))\n",
-      "/coding/yolort/yolort/models/transform.py:282: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).\n",
+      "/coding/yolov5-rt-stack/yolort/models/transform.py:282: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).\n",
       "  img_h, img_w = _get_shape_onnx(img)\n",
-      "/coding/yolort/yolort/models/anchor_utils.py:45: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
+      "/coding/yolov5-rt-stack/yolort/models/anchor_utils.py:45: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
       "  anchors = torch.as_tensor(self.anchor_grids, dtype=torch.float32, device=device).to(dtype=dtype)\n",
-      "/coding/yolort/yolort/models/anchor_utils.py:46: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
+      "/coding/yolov5-rt-stack/yolort/models/anchor_utils.py:46: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
       "  strides = torch.as_tensor(self.strides, dtype=torch.float32, device=device).to(dtype=dtype)\n",
-      "/coding/yolort/yolort/models/box_head.py:402: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
+      "/coding/yolov5-rt-stack/yolort/models/box_head.py:402: TracerWarning: torch.as_tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.\n",
       "  strides = torch.as_tensor(self.strides, dtype=torch.float32, device=device).to(dtype=dtype)\n",
-      "/coding/yolort/yolort/models/box_head.py:333: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).\n",
+      "/coding/yolov5-rt-stack/yolort/models/box_head.py:333: TracerWarning: Iterating over a tensor might cause the trace to be incorrect. Passing a tensor of different shape won't change the number of iterations executed (and might lead to errors or silently give incorrect results).\n",
       "  for head_output, grid, shift, stride in zip(head_outputs, grids, shifts, strides):\n",
       "/opt/conda/lib/python3.8/site-packages/torch/onnx/symbolic_opset9.py:2815: UserWarning: Exporting aten::index operator of advanced indexing in opset 11 is achieved by combination of multiple ONNX operators, including Reshape, Transpose, Concat, and Gather. If indices include negative values, the exported graph will produce incorrect results.\n",
       "  warnings.warn(\"Exporting aten::index operator of advanced indexing in opset \" +\n"