Some vertical domain recognition tasks (e.g., vehicles, commodities, etc.) require a large number of recognized categories, and often use a retrieval-based approach to obtain matching predicted categories by performing a fast nearest neighbor search with query vectors and underlying library vectors. The vector search module provides the basic approximate nearest neighbor search algorithm based on Baidu's self-developed Möbius algorithm, a graph-based approximate nearest neighbor search algorithm for maximum inner product search (MIPS). This module provides python interface, supports numpy and tensor type vectors, and supports L2 and Inner Product distance calculation.
Details of the Mobius algorithm can be found in the paper.（[Möbius Transformation for Fast Inner Product Search on Graph](http://research.baidu.com/Public/uploads/5e189d36b5cf6.PDF), [Code](https://github.com/sunbelbd/mobius)）
## 2. Installation
### 2.1 Use the provided library files directly
This folder contains the compiled `index.so` (compiled under gcc8.2.0 for Linux) and `index.dll` (compiled under gcc10.3.0 for Windows), which can be used directly, skipping sections 2.2 and 2.3.
If the library files are not available due to a low gcc version or an incompatible environment, you need to manually compile the library files under a different platform.
**Note：** Make sure that C++ compiler supports the C++11 standard.
### 2.2 Compile and generate library files on Linux
Run the following command to install gcc and g++.
```
sudo apt-get update
sudo apt-get upgrade -y
sudo apt-get install build-essential gcc g++
```
Check the gcc version by the command `gcc -v`.
`make` can be operated directly. If you wish to regenerate the `index.so`, you can first use `make clean` to clear the cache, and then use `make` to generate the updated library file.
### 2.3 Compile and generate library files on Windows
You need to install gcc compiler tool first, we recommend using [TDM-GCC](https://jmeubank.github.io/tdm-gcc/articles/2020-03/9.2.0-release), you can choose the right version on the official website. We recommend downloading [tdm64-gcc-10.3.0-2.exe](https://github.com/jmeubank/tdm-gcc/releases/download/v10.3.0-tdm64-2/tdm64-gcc-10.3.0-2.exe).
After the downloading, follow the default installation steps to install. There are 3 points to note here:
1. The vector search module depends on openmp, so you need to check the `openmp` installation option when going on to `choose components` step, otherwise it will report an error `libgomp.spec: No such file or directory`, [reference link](https://github.com/dmlc/xgboost/issues/1027)
2. When being asked whether to add to the system environment variables, it is recommended to check here, otherwise you need to add the system environment variables manually later.
3. The compile command is `make` on Linux and `mingw32-make` on Windows, so you need to distinguish here.
After installation, you can open a command line terminal and check the gcc version with the command `gcc -v`.
Run the command `mingw32-make` to generate the `index.dll` library file under the folder (deploy/vector_search). If you want to regenerate the `index.dll` file, you can first use `mingw32-make clean` to clear the cache, and then use `mingw32-make` to generate the updated library file.
### 2.4 Compile and generate library files on MacOS
Run the following command to install gcc and g++:
```
brew install gcc
```
#### Caution：
1. If prompted with ``Error: Running Homebrew as root is extremely dangerous and no longer supported... `, refer to this [link](https://jingyan.baidu.com/article/e52e3615057a2840c60c519c.html)
2. If prompted with `Error: Failure while executing; `tar --extract --no-same-owner --file... ', refer to this [link](https://blog.csdn.net/Dawn510/article/details/117787358).
After installation the compiled executable is copied under /usr/local/bin, look at the gcc in this folder:
```
ls /usr/local/bin/gcc*
```
The local gcc version is gcc-11, and the compile command is as follows: (If the local gcc version is gcc-9, the corresponding command should be `CXX=g++-9 make`)
The parameters in the PaddleClas configuration file(`ppcls/configs/*.yaml`)are described for you to customize or modify the hyperparameter configuration more quickly.
## Details
### 1. Classification model
Here the configuration of `ResNet50_vd` on`ImageNet-1k`is used as an example to explain the each parameter in detail. [Configure Path](https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/configs/ImageNet/ResNet/ResNet50_vd.yaml).
#### 1.1Global Configuration
| Parameter name | Specific meaning | Defult value | Optional value |
**Note**: Here pretrained can be set to True or False, so does the path of the weights. In addition, the pretrained is disabled when Global.pretrained_model is also set to the corresponding path.
#### 1.3 Loss function
| Parameter name | Specific meaning | Defult value | Optional value |
**Note**：The new parameters may be different when `lr.name` is different , as when `lr.name=Piecewise`, the following parameters need to be added:
```
lr:
name: Piecewise
learning_rate: 0.1
decay_epochs: [30, 60, 90]
values: [0.1, 0.01, 0.001, 0.0001]
```
Referring to [learning_rate.py](https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/optimizer/learning_rate.py) for adding method and parameters.
#### 1.5 Data reading module（DataLoader）
##### 1.5.1 dataset
| Parameter name | Specific meaning | Defult value | Optional value |
| infer_imgs | Image address to be inferred | docs/images/whl/demo.jpg | str |
| batch_size | batch size | 10 | int |
| PostProcess.name | Post-process name | Topk | str |
| PostProcess.topk | topk value | 5 | int |
| PostProcess.class_id_map_file | mapping file of class id and name | ppcls/utils/imagenet1k_label_list.txt | str |
**Note**：The interpretation of `transforms` in the Infer module refers to the interpretation of`transform_ops`in the dataset in the data reading module.
### 2.Distillation model
**Note**：Here the training configuration of `MobileNetV3_large_x1_0` on `ImageNet-1k` distilled MobileNetV3_small_x1_0 is used as an example to explain the meaning of each parameter in detail. [Configure path](https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/configs/ImageNet/Distillation/mv3_large_x1_0_distill_mv3_small_x1_0.yaml). Only parameters that are distinct from the classification model are introduced here.
#### 2.1 Architecture
| Parameter name | Specific meaning | Default meaning | Optional meaning |
| DistillationTopkAcc | DistillationTopkAcc | including model_key and topk | —— |
| DistillationTopkAcc.model_key | the evaluated model | "Student" | "Teacher" |
| DistillationTopkAcc.topk | Topk value | [1, 5] | list, int |
**Note**： `DistillationTopkAcc` has the same meaning as `TopkAcc`, except that it is only used in distillation tasks.
### 3. Recognition model
**Note**：The training configuration of`ResNet50` on`LogoDet-3k` is used here as an example to explain the meaning of each parameter in detail. [configure path](https://github.com/PaddlePaddle/PaddleClas/blob/develop/ppcls/configs/Logo/ResNet50_ReID.yaml). Only parameters that are distinct from the classification model are presented here.
#### 3.1 Architechture
| Parameter name | Specific meaning | Default meaning | Optional meaning |
| infer_add_softmax | softmaxwhether to add softmax to infercne | False | [True, False] |
| Backbone.name | Backbone name | ResNet50_last_stage_stride1 | other backbone provided by PaddleClas |
| Backbone.pretrained | Backbone pre-trained model | True | Boolean value or pre-trained model path |
| BackboneStopLayer.name | The name of the output layer in Backbone | True | The`full_name`of the feature output layer in Backbone |
| Neck.name | The name of the Neck part | VehicleNeck | the dictionary structure to be passed in, the specific input parameters for the Neck network layer |
| Neck.in_channels | Input dimension size of the Neck part | 2048 | the size is the same as BackboneStopLayer.name |
| Neck.out_channels | Output the dimension size of the Neck part, i.e. feature dimension size | 512 | int |
| Head.name | Network Head part nam | CircleMargin | Arcmargin. Etc |
| Head.class_num | number of classes | 3000 | int |
| Head.margin | margin value in CircleMargin | 0.35 | float |
| Head.scale | scale value in CircleMargin | 64 | int |
**Note**：
1.In PaddleClas, the `Neck` part is the connection part between Backbone and embedding layer, and `Head` part is the connection part between embedding layer and classification layer.。
2.`BackboneStopLayer.name can be obtained by visualizing the model, visualization can be referred to [Netron](https://github.com/lutzroeder/netron)或者[visualdl](https://github.com/PaddlePaddle/VisualDL).
3.Calling tools/export_model.py will convert the model weights to inference model, where the infer_add_softmax parameter will control whether to add the Softmax activation function afterwards, the code default is True (the last output layer in the classification task will be connected to the Softmax activation function). In the recognition task, the activation function is not required for the feature layer, so it should be set to False here.
#### 3.2 Evaluation metric
| Parameter name | Specific meaning | Default meaning | Optional meaning |
This document introduces the configuration(filed in `config/*.yaml`) of PaddleClas.
* Note: Some parameters do not appear in the yaml file (because they are not used for this file). During training or validation, you can use the command `-o` to update or add the specified parameters. For the example `-o checkpoints=./ckp_path/ppcls`, it means that the parameter `checkpoints` will be updated or added using the value `./ckp_path/ppcls`.
### Basic
| name | detail | default value | optional value |
|:---:|:---:|:---:|:---:|
| mode | mode | "train" | ["train"," valid"] |
| checkpoints | checkpoint model path for resuming training process | "" | Str |
| last_epoch | last epoch for the training，used with checkpoints | -1 | int |
