support yolov6lite inference

configs/yolov6/yolov6lite/README.md

@@ -0,0 +1,14 @@
+# YOLOv6Lite
+
+## 模型库
+### YOLOv6Lite on COCO
+
+| 网络网络        | 输入尺寸   | 图片数/GPU | 学习率策略 | TRT-FP16-Latency(ms) |   mAP  |   AP50  | Params(M) | FLOPs(G) |  下载链接       | 配置文件 |
+| :------------- | :------- | :-------: | :--------------: | :---------: | :-----: |:-----: | :-----: |:-----: | :-------------: | :-----: |
+| *YOLOv6Lite-s  |  320     |    32      |   400e |  -  |  22.3 | 34.2 |  0.55  | 0.56 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6lite_s_400e_coco.pdparams) | [配置文件](./yolov6lite_s_400e_coco.yml) |
+| *YOLOv6Lite-m  |  320     |    32      |   400e |  -  |  24.8 | 37.7 |  0.79  | 0.67 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6lite_m_400e_coco.pdparams) | [配置文件](./yolov6lite_m_400e_coco.yml) |
+| *YOLOv6Lite-l  |  320     |    32      |   400e |  -  |  27.6 | 41.6 |  1.09  | 0.87 |[下载链接](https://paddledet.bj.bcebos.com/models/yolov6lite_l_400e_coco.pdparams) | [配置文件](./yolov6lite_l_400e_coco.yml) |
+
+
+**注意:**
+  - YOLOv6Lite 模型暂未支持完全训练；训练使用COCO train2017作为训练集，mAP为在COCO val2017上的`mAP(IoU=0.5:0.95)`结果，且评估未使用`multi_label`等trick；

configs/yolov6/yolov6lite/_base_/optimizer_400e.yml

@@ -0,0 +1,19 @@
+epoch: 400
+
+LearningRate:
+  base_lr: 0.01
+  schedulers:
+  - !YOLOv5LRDecay
+    max_epochs: 400
+    min_lr_ratio: 0.01
+  - !ExpWarmup
+    epochs: 3
+
+OptimizerBuilder:
+  optimizer:
+    type: Momentum
+    momentum: 0.937
+    use_nesterov: True
+  regularizer:
+    factor: 0.0005
+    type: L2

configs/yolov6/yolov6lite/_base_/optimizer_400e_fintune.yml

@@ -0,0 +1,19 @@
+epoch: 400
+
+LearningRate:
+  base_lr: 0.0032
+  schedulers:
+  - !YOLOv5LRDecay
+    max_epochs: 400
+    min_lr_ratio: 0.12
+  - !ExpWarmup
+    epochs: 2
+
+OptimizerBuilder:
+  optimizer:
+    type: Momentum
+    momentum: 0.843
+    use_nesterov: True
+  regularizer:
+    factor: 0.00036
+    type: L2

configs/yolov6/yolov6lite/_base_/yolov6lite_effibackbone.yml

@@ -0,0 +1,48 @@
+architecture: YOLOv6
+norm_type: sync_bn
+use_ema: True
+ema_decay: 0.9999
+ema_decay_type: "exponential"
+find_unused_parameters: True
+
+act: 'relu'
+training_mode: "repvgg"
+self_distill: False
+
+width_mult: 1.0
+
+YOLOv6:
+  backbone: Lite_EffiBackbone
+  neck: Lite_EffiNeck
+  yolo_head: Lite_EffideHead
+  post_process: ~
+
+Lite_EffiBackbone:
+  return_idx: [2, 3, 4]
+
+Lite_EffiNeck:
+  unified_channels: 96
+
+Lite_EffideHead:
+  fpn_strides: [8, 16, 32, 64]
+  grid_cell_scale: 5.0
+  grid_cell_offset: 0.5
+  reg_max: 0
+  use_dfl: False
+  static_assigner_epoch: 4 # warmup_epoch
+  loss_weight: {cls: 1.0, iou: 2.5}
+  iou_type: 'siou' # 'siou' in lite s/m/l
+  static_assigner:
+    name: ATSSAssigner
+    topk: 9
+  assigner:
+    name: TaskAlignedAssigner
+    topk: 13
+    alpha: 1.0
+    beta: 6.0
+  nms:
+    name: MultiClassNMS
+    nms_top_k: 2000
+    keep_top_k: 300
+    score_threshold: 0.03
+    nms_threshold: 0.65

configs/yolov6/yolov6lite/_base_/yolov6lite_reader_320.yml

@@ -0,0 +1,52 @@
+input_height: &input_height 320
+input_width: &input_width 320
+input_size: &input_size [*input_height, *input_width]
+mosaic_epoch: &mosaic_epoch 400
+
+worker_num: 4
+TrainReader:
+  sample_transforms:
+    - Decode: {}
+    - MosaicPerspective:
+        mosaic_prob: 1.0
+        target_size: *input_size
+        degrees: 0.0
+        translate: 0.1
+        scale: 0.5
+        shear: 0.0
+        mixup_prob: 0.0
+    - RandomHSV: {hgain: 0.015, sgain: 0.7, vgain: 0.4}
+    - RandomFlip: {}
+  batch_transforms:
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - Permute: {}
+    - PadGT: {}
+  batch_size: 32
+  shuffle: True
+  drop_last: True
+  use_shared_memory: True
+  collate_batch: True
+  mosaic_epoch: *mosaic_epoch
+
+
+EvalReader:
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
+    - Pad: {size: *input_size, fill_value: [114., 114., 114.]}
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - Permute: {}
+  batch_size: 1
+
+
+TestReader:
+  inputs_def:
+    image_shape: [3, *input_height, *input_width]
+  sample_transforms:
+    - Decode: {}
+    - Resize: {target_size: *input_size, keep_ratio: True, interp: 1}
+    - Pad: {size: *input_size, fill_value: [114., 114., 114.]}
+    - NormalizeImage: {mean: [0., 0., 0.], std: [1., 1., 1.], norm_type: none}
+    - Permute: {}
+  batch_size: 1
+  fuse_normalize: False

configs/yolov6/yolov6lite/yolov6lite_l_400e_coco.yml

@@ -0,0 +1,19 @@
+_BASE_: [
+  '../../datasets/coco_detection.yml',
+  '../../runtime.yml',
+  '_base_/optimizer_400e.yml',
+  '_base_/yolov6lite_effibackbone.yml',
+  '_base_/yolov6lite_reader_320.yml',
+]
+width_mult: 1.5
+log_iter: 100
+snapshot_epoch: 10
+weights: output/yolov6lite_l_400e_coco/model_final
+
+
+### reader config
+TrainReader:
+  batch_size: 32 # default 8 gpus, total bs = 256
+
+EvalReader:
+  batch_size: 8

configs/yolov6/yolov6lite/yolov6lite_m_400e_coco.yml

@@ -0,0 +1,19 @@
+_BASE_: [
+  '../../datasets/coco_detection.yml',
+  '../../runtime.yml',
+  '_base_/optimizer_400e.yml',
+  '_base_/yolov6lite_effibackbone.yml',
+  '_base_/yolov6lite_reader_320.yml',
+]
+width_mult: 1.1
+log_iter: 100
+snapshot_epoch: 10
+weights: output/yolov6lite_m_400e_coco/model_final
+
+
+### reader config
+TrainReader:
+  batch_size: 32 # default 8 gpus, total bs = 256
+
+EvalReader:
+  batch_size: 8

configs/yolov6/yolov6lite/yolov6lite_s_400e_coco.yml

@@ -0,0 +1,19 @@
+_BASE_: [
+  '../../datasets/coco_detection.yml',
+  '../../runtime.yml',
+  '_base_/optimizer_400e.yml',
+  '_base_/yolov6lite_effibackbone.yml',
+  '_base_/yolov6lite_reader_320.yml',
+]
+width_mult: 0.7
+log_iter: 100
+snapshot_epoch: 10
+weights: output/yolov6lite_s_400e_coco/model_final
+
+
+### reader config
+TrainReader:
+  batch_size: 32 # default 8 gpus, total bs = 256
+
+EvalReader:
+  batch_size: 8