update readme and config files for base model

README.md

@@ -28,10 +28,10 @@ YOLOv6-N hits 35.9% AP on COCO dataset with 1234 FPS on T4. YOLOv6-S strikes 43.
 
 ## What's New
 
-- Release M/L models and update N/T/S models with enhanced performance.⭐️ [Benchmark](#Benchmark)
-- 2x faster training time.
-- Fix the degration of performance when evaluating on 640x640 inputs.
-- Customized quantization methods. 🚀 [Quantization Tutorial](./tools/qat/README.md)
+- [2022.11.04] Release [base models](configs/base/README.md) to simplify the training and deployment process.
+- [2022.09.06] Customized quantization methods. 🚀 [Quantization Tutorial](./tools/qat/README.md)
+- [2022.09.05] Release M/L models and update N/T/S models with enhanced performance.⭐️ [Benchmark](#Benchmark)
+- [2022.06.23] Release N/T/S models with excellent performance.
 
 ## Benchmark
 | Model                                                        | Size | mAP<sup>val<br/>0.5:0.95              | Speed<sup>T4<br/>trt fp16 b1 <br/>(fps) | Speed<sup>T4<br/>trt fp16 b32 <br/>(fps) | Params<br/><sup> (M) | FLOPs<br/><sup> (G) |

README_cn.md

@@ -16,12 +16,12 @@ YOLOv6 提供了一系列面向各种工业应用场景的模型，包括微小
 
 YOLOv6-N 在 COCO 数据集上的 mAP 为 35.9% ，在 T4 显卡推理速度可达 1234 FPS。 YOLOv6-S 的 mAP 为 43.5%，推理速度为 495 FPS ，量化后的 YOLOv6-S 模型在 T4显卡 上的 FPS 可以加速到 869，mAP 为 43.3% 。 YOLOv6-T/M/L 也具有出色的性能，与其他检测器相比，我们的模型在基本相同的推理速度时，可以达到更高的精度。
 
-## 本次更新
+## 更新日志
 
-- 发布了 M/L 模型，并且进一步提高了 N/T/S 模型的性能⭐️ [精度指标](#模型指标) 
-- 将训练速度提高了2倍；
-- 修复了 640x640 分辨率推理时性能下降的问题；
-- 定制化的模型量化加速方法 🚀 [量化教程](./tools/qat/README.md)；
+- [2022.11.04] 发布 [基础版模型](configs/base/README_cn.md) 简化训练部署流程。
+- [2022.09.06] 定制化的模型量化加速方法 🚀 [量化教程](./tools/qat/README.md)
+- [2022.09.05] 发布 M/L 模型，并且进一步提高了 N/T/S 模型的性能 ⭐️ [模型指标](#模型指标)
+- [2022.06.23] 发布 N/T/S v1.0 版本模型。
 
 ## 模型指标
 | 模型                                                       | 输入尺寸 | mAP<sup>val<br/>0.5:0.95              | 速度<sup>T4<br/>trt fp16 b1 <br/>(fps) | 速度<sup>T4<br/>trt fp16 b32 <br/>(fps) | Params<br/><sup> (M) | FLOPs<br/><sup> (G) |

configs/base/README.md

@@ -0,0 +1,28 @@
+## YOLOv6 base model
+
+English | [简体中文](./README_cn.md)
+
+### Features
+
+- Use only regular convolution and Relu activation functions.
+
+- Apply CSP (1/2 channel dim) blocks in the network structure, except for Nano base model.
+
+Advantage:
+- Adopt a unified network structure and configuration, and the accuracy loss of the PTQ 8-bit quantization model is negligible, about 0.4%.
+- Suitable for users who are just getting started or who need to apply, optimize and deploy an 8-bit quantization model quickly and frequently.
+
+Shortcoming:
+- The accuracy on COCO is slightly lower than the v2.0 released models.
+
+### Performance
+
+| Model                                                         | Size | mAP<sup>val<br/>0.5:0.95 | Speed<sup>T4<br/>trt fp16 b1 <br/>(fps) | Speed<sup>T4<br/>trt fp16 b32 <br/>(fps) | Params<br/><sup> (M) | FLOPs<br/><sup> (G) |
+| :----------------------------------------------------------- | -------- | :----------------------- | -------------------------------------- | --------------------------------------- | -------------------- | ------------------- |
+| [**YOLOv6-N-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6n_base.pt) | 640      | 35.6<sup>400e            | 832                                    | 1249                                    | 4.3                  | 11.1                |
+| [**YOLOv6-S-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6s_base.pt) | 640      | 43.8<sup>400e            | 373                                    | 531                                     | 11.5                 | 27.6                |
+| [**YOLOv6-M-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6m_base.pt) | 640      | 48.8<sup>distill         | 179                                    | 246                                     | 27.7                 | 68.4                |
+| [**YOLOv6-L-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6l_base.pt) | 640      | 51.0<sup>distill         | 115                                    | 153                                     | 58.5                 | 144.0               |
+
+- Speed is tested with TensorRT 7.2 on T4.
+- The processes of model training, evaluation, and inference are the same as the original ones. For details, please refer to [this README](https://github.com/meituan/YOLOv6#quick-start).

configs/base/README_cn.md

@@ -0,0 +1,27 @@
+## YOLOv6 基础版模型
+
+简体中文 | [English](./README.md)
+
+### 模型特点
+
+- 仅使用常规卷积和Relu激活函数
+
+- 网络结构均采用CSP (1/2通道) block，Nano网络除外。
+
+优势：
+- 采用统一的网络结构和配置，且 PTQ 8位量化模型精度损失(约0.4%)较小，适合刚入门或有快速迭代部署8位量化模型需求的用户。
+
+不足：
+- COCO上精度对比2.0版本发布模型稍低。
+
+### 模型指标
+
+| 模型                                                         | 输入尺寸 | mAP<sup>val<br/>0.5:0.95 | 速度<sup>T4<br/>trt fp16 b1 <br/>(fps) | 速度<sup>T4<br/>trt fp16 b32 <br/>(fps) | Params<br/><sup> (M) | FLOPs<br/><sup> (G) |
+| :----------------------------------------------------------- | -------- | :----------------------- | -------------------------------------- | --------------------------------------- | -------------------- | ------------------- |
+| [**YOLOv6-N-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6n_base.pt) | 640      | 35.6<sup>400e            | 832                                    | 1249                                    | 4.3                  | 11.1                |
+| [**YOLOv6-S-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6s_base.pt) | 640      | 43.8<sup>400e            | 373                                    | 531                                     | 11.5                 | 27.6                |
+| [**YOLOv6-M-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6m_base.pt) | 640      | 48.8<sup>distill         | 179                                    | 246                                     | 27.7                 | 68.4                |
+| [**YOLOv6-L-base**](https://github.com/meituan/YOLOv6/releases/download/0.2.1/yolov6l_base.pt) | 640      | 51.0<sup>distill         | 115                                    | 153                                     | 58.5                 | 144.0               |
+
+- 速度是在 T4 上测试的，TensorRT 版本为 7.2；
+- 模型训练、评估、推理流程与原来保持一致，具体可参考 [首页 README 文档](https://github.com/meituan/YOLOv6/blob/main/README_cn.md#%E5%BF%AB%E9%80%9F%E5%BC%80%E5%A7%8B)。

configs/base/yolov6l_base.py

@@ -0,0 +1,62 @@
+# YOLOv6 large base model
+model = dict(
+    type='YOLOv6l_base',
+    depth_multiple=1.0,
+    width_multiple=1.0,
+    pretrained=None,
+    backbone=dict(
+        type='CSPBepBackbone',
+        num_repeats=[1, 6, 12, 18, 6],
+        out_channels=[64, 128, 256, 512, 1024],
+        csp_e=float(1)/2,
+        ),
+    neck=dict(
+        type='CSPRepPANNeck',
+        num_repeats=[12, 12, 12, 12],
+        out_channels=[256, 128, 128, 256, 256, 512],
+        csp_e=float(1)/2,
+        ),
+    head=dict(
+        type='EffiDeHead',
+        in_channels=[128, 256, 512],
+        num_layers=3,
+        begin_indices=24,
+        anchors=1,
+        out_indices=[17, 20, 23],
+        strides=[8, 16, 32],
+        iou_type='giou',
+        use_dfl=True,
+        reg_max=16, #if use_dfl is False, please set reg_max to 0
+        distill_weight={
+            'class': 2.0,
+            'dfl': 1.0,
+        },
+    )
+)
+
+solver=dict(
+    optim='SGD',
+    lr_scheduler='Cosine',
+    lr0=0.01,
+    lrf=0.01,
+    momentum=0.937,
+    weight_decay=0.0005,
+    warmup_epochs=3.0,
+    warmup_momentum=0.8,
+    warmup_bias_lr=0.1       
+)
+
+data_aug = dict(
+    hsv_h=0.015,  
+    hsv_s=0.7, 
+    hsv_v=0.4,
+    degrees=0.0,
+    translate=0.1,
+    scale=0.9,
+    shear=0.0,
+    flipud=0.0,
+    fliplr=0.5,
+    mosaic=1.0,
+    mixup=0.1,
+)
+training_mode = "conv_relu" 

configs/base/yolov6l_base_finetune.py

@@ -0,0 +1,62 @@
+# YOLOv6 large base model
+model = dict(
+    type='YOLOv6l_base',
+    depth_multiple=1.0,
+    width_multiple=1.0,
+    pretrained=None,
+    backbone=dict(
+        type='CSPBepBackbone',
+        num_repeats=[1, 6, 12, 18, 6],
+        out_channels=[64, 128, 256, 512, 1024],
+        csp_e=float(1)/2,
+        ),
+    neck=dict(
+        type='CSPRepPANNeck',
+        num_repeats=[12, 12, 12, 12],
+        out_channels=[256, 128, 128, 256, 256, 512],
+        csp_e=float(1)/2,
+        ),
+    head=dict(
+        type='EffiDeHead',
+        in_channels=[128, 256, 512],
+        num_layers=3,
+        begin_indices=24,
+        anchors=1,
+        out_indices=[17, 20, 23],
+        strides=[8, 16, 32],
+        iou_type='giou',
+        use_dfl=True,
+        reg_max=16, #if use_dfl is False, please set reg_max to 0
+        distill_weight={
+            'class': 2.0,
+            'dfl': 1.0,
+        },
+    )
+)
+
+solver = dict(
+    optim='SGD',
+    lr_scheduler='Cosine',
+    lr0=0.0032,
+    lrf=0.12,
+    momentum=0.843,
+    weight_decay=0.00036,
+    warmup_epochs=2.0,
+    warmup_momentum=0.5,
+    warmup_bias_lr=0.05
+)
+
+data_aug = dict(
+    hsv_h=0.0138,
+    hsv_s=0.664,
+    hsv_v=0.464,
+    degrees=0.373,
+    translate=0.245,
+    scale=0.898,
+    shear=0.602,
+    flipud=0.00856,
+    fliplr=0.5,
+    mosaic=1.0,
+    mixup=0.243,
+)
+training_mode = "conv_relu" 

configs/base/yolov6m_base.py

@@ -0,0 +1,62 @@
+# YOLOv6 medium/large base model
+model = dict(
+    type='YOLOv6m_base',
+    depth_multiple=0.80,
+    width_multiple=0.75,
+    pretrained=None,
+    backbone=dict(
+        type='CSPBepBackbone',
+        num_repeats=[1, 6, 12, 18, 6],
+        out_channels=[64, 128, 256, 512, 1024],
+        csp_e=float(1)/2,
+        ),
+    neck=dict(
+        type='CSPRepPANNeck',
+        num_repeats=[12, 12, 12, 12],
+        out_channels=[256, 128, 128, 256, 256, 512],
+        csp_e=float(1)/2,
+        ),
+    head=dict(
+        type='EffiDeHead',
+        in_channels=[128, 256, 512],
+        num_layers=3,
+        begin_indices=24,
+        anchors=1,
+        out_indices=[17, 20, 23],
+        strides=[8, 16, 32],
+        iou_type='giou',
+        use_dfl=True,
+        reg_max=16, #if use_dfl is False, please set reg_max to 0
+        distill_weight={
+            'class': 1.0,
+            'dfl': 1.0,
+        },
+    )
+)
+
+solver=dict(
+    optim='SGD',
+    lr_scheduler='Cosine',
+    lr0=0.01,
+    lrf=0.01,
+    momentum=0.937,
+    weight_decay=0.0005,
+    warmup_epochs=3.0,
+    warmup_momentum=0.8,
+    warmup_bias_lr=0.1       
+)
+
+data_aug = dict(
+    hsv_h=0.015,  
+    hsv_s=0.7, 
+    hsv_v=0.4,
+    degrees=0.0,
+    translate=0.1,
+    scale=0.9,
+    shear=0.0,
+    flipud=0.0,
+    fliplr=0.5,
+    mosaic=1.0,
+    mixup=0.1,
+)
+training_mode = "conv_relu" 

configs/base/yolov6m_base_finetune.py

@@ -0,0 +1,62 @@
+# YOLOv6 medium/large base model
+model = dict(
+    type='YOLOv6m_base',
+    depth_multiple=0.80,
+    width_multiple=0.75,
+    pretrained=None,
+    backbone=dict(
+        type='CSPBepBackbone',
+        num_repeats=[1, 6, 12, 18, 6],
+        out_channels=[64, 128, 256, 512, 1024],
+        csp_e=float(1)/2,
+        ),
+    neck=dict(
+        type='CSPRepPANNeck',
+        num_repeats=[12, 12, 12, 12],
+        out_channels=[256, 128, 128, 256, 256, 512],
+        csp_e=float(1)/2,
+        ),
+    head=dict(
+        type='EffiDeHead',
+        in_channels=[128, 256, 512],
+        num_layers=3,
+        begin_indices=24,
+        anchors=1,
+        out_indices=[17, 20, 23],
+        strides=[8, 16, 32],
+        iou_type='giou',
+        use_dfl=True,
+        reg_max=16, #if use_dfl is False, please set reg_max to 0
+        distill_weight={
+            'class': 1.0,
+            'dfl': 1.0,
+        },
+    )
+)
+
+solver = dict(
+    optim='SGD',
+    lr_scheduler='Cosine',
+    lr0=0.0032,
+    lrf=0.12,
+    momentum=0.843,
+    weight_decay=0.00036,
+    warmup_epochs=2.0,
+    warmup_momentum=0.5,
+    warmup_bias_lr=0.05
+)
+
+data_aug = dict(
+    hsv_h=0.0138,
+    hsv_s=0.664,
+    hsv_v=0.464,
+    degrees=0.373,
+    translate=0.245,
+    scale=0.898,
+    shear=0.602,
+    flipud=0.00856,
+    fliplr=0.5,
+    mosaic=1.0,
+    mixup=0.243,
+)
+training_mode = "conv_relu" 

configs/base/yolov6n_base.py

@@ -0,0 +1,56 @@
+# YOLOv6 nano base model
+model = dict(
+    type='YOLOv6n_base',
+    depth_multiple=0.33,
+    width_multiple=0.25,
+    pretrained=None,
+    backbone=dict(
+        type='EfficientRep',
+        num_repeats=[1, 6, 12, 18, 6],
+        out_channels=[64, 128, 256, 512, 1024],
+        ),
+    neck=dict(
+        type='RepPANNeck',
+        num_repeats=[12, 12, 12, 12],
+        out_channels=[256, 128, 128, 256, 256, 512],
+        ),
+    head=dict(
+        type='EffiDeHead',
+        in_channels=[128, 256, 512],
+        num_layers=3,
+        begin_indices=24,
+        anchors=1,
+        out_indices=[17, 20, 23],
+        strides=[8, 16, 32],
+        iou_type='giou',
+        use_dfl=False,
+        reg_max=0 #if use_dfl is False, please set reg_max to 0
+    )
+)
+
+solver=dict(
+    optim='SGD',
+    lr_scheduler='Cosine',
+    lr0=0.01,
+    lrf=0.01,
+    momentum=0.937,
+    weight_decay=0.0005,
+    warmup_epochs=3.0,
+    warmup_momentum=0.8,
+    warmup_bias_lr=0.1       
+)
+
+data_aug = dict(
+    hsv_h=0.015,  
+    hsv_s=0.7, 
+    hsv_v=0.4,
+    degrees=0.0,
+    translate=0.1,
+    scale=0.5,
+    shear=0.0,
+    flipud=0.0,
+    fliplr=0.5,
+    mosaic=1.0,
+    mixup=0.0,
+)
+training_mode = "conv_relu"