AISVision AI Software Toolkit User Manual v1.3.1 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . About this manual ............................................................................... 4 Chapter 1: Getting to know AISVision .................................................. 7 1.1 Introduction .................................................................................................... 7 1.1.1 Features: .................................................................................................. 7 1.1.2 Trainer...........................................................
ASUS, Inc . 2.2.3.2 Create and delete task(s).....................................................................37 2.2.3.3 How to modify task detail ...................................................................37 2.2.3.4 Task detail confirmation......................................................................38 2.2.3.5 Edit task detail ....................................................................................39 2.2.3.6 Start training...................................................
ASUS, Inc . 2.2.6.2 Scheduling ..........................................................................................57 2.2.6.3 Re-arrange parameter(s) .....................................................................57 2.2.6.4 Start training.......................................................................................58 Chapter 3: Upgrading AISVision.......................................................... 59 3.1 Complimentary AISVision upgrade policy....................................
ASUS, Inc . About this manual This manual provides an overview of the features of this AI software toolkit and gives step-by-step instructions for making full use of them. Release notes Version 1.0 Description Modify by Date Eric 2021/08/18 Xu 2021/12/22 Wayne 2022/06/14 1. Modified index 2. Add Intel® OpenVINO support in 2.2.4.10 3. Add VGA card QVL Eric 2022/07/29 1. Updated QVL Eric 2022/10/17 1.First released version 2.Set 2.2.1 to 2.3.4 as help content 3.Set 1.1 and 1.1.1 as about 1.
ASUS, Inc . How this manual is organized Chapter 1 Chapter 4 Getting to know AISVision This chapter details the key features and functions of this AI toolkit. Using AISVision This chapter provides information on how to use this AI toolkit. Upgrading AISVision This chapter provides information on how to upgrade the toolkit version, pre-trained models, and other necessary functions. Support for AISVision Appendix This chapter provides information for troubleshooting and contacting ASUS for support.
ASUS, Inc . Conventions used in this manual Throughout this manual, blocks of text as shown below are used to emphasize important information in this manual. IMPORTANT! This message contains vital information that must be followed to complete a task. NOTE: This message contains additional information and tips that can help to complete a task.
ASUS, Inc . Chapter 1: Getting to know AISVision 1.1 Introduction AISVision is an advanced AI training tool with a unique and intuitive user interface that allows you to quickly train and easily deploy AI models in the field. With four (4) vision AI functions (Abnormal detection, Fast object detection, Fast classification, and Fast anomaly detection), this all-in-one tool can quickly and efficiently detect and recognize objects, defects, and other characteristics in the field. 1.1.
ASUS, Inc . 1.2 About the AISVision Support API AISVision now supports C++/C/C# for extended development. For a developer guide for programming find the API reference file on ASUS IoT website under the Driver & Utility of Technical Resources. https://iot.asus.com/products/AI-software/AISVision/ 1.3 System Requirements Your computer must meet the minimum system requirements below to run the artificial intelligence model training and inference functions in this software toolkit.
ASUS, Inc . Chapter 2: Using AISVision 2.1 Download the toolkit Go to the ASUS IoT website and find the AISVision product download page 2.1.2 Install the toolkit Run installation.exe once the image file is downloaded. 2.1.3 Activate the toolkit When prompted for the activation key, plug the ASUS IoT USB dongle into the computer that you have installed AISVision and can use it to perform the product quality inspection or object recognition. 2.
ASUS, Inc . 2.2.1 Project In this section, you can set up a project to define and keep information, including desired features and functions, for use in subsequent steps. Create a new project Import project(s) Open a project Add/modify description Description in project detail box 2.2.1.1 Create a new project Follow the top row, enter a new project name, follow by desired function this project need to be performed, confirm storage path, click "ADD".
ASUS, Inc . 2.2.1.2 Import project(s) Click "Import" button to popup a dialog window, double click or select and click "import". *.ditprj sub-file name only. 2.2.1.3 Open a project Load a project via process below : 1. Click project icon (green box) to load a project. 2. Double click any space on the screen to load a project. Other: 1. 2. When project loaded, modify/delete etc function keys shown. When project loaded, a blue bar shown alongside the project detail information.
ASUS, Inc . 2.2.1.4 Add/modify description Tap on the description box to add or modify the project description, and click "Save" icon to save the change. * Description changed will be shown in project detail below and on the top of the project title. 2.2.1.
ASUS, Inc . 2.2.2 Labeller This function allows you to load and label images for the next steps to complete this AI function and has four (4) categories: segmentation, object detection, classification, and anomaly detection as shown below. 13 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.1 Labeller - Segmentation Allow users to label the required message in the images for Segmentation process. Add category Add image(s) Labeling Modify lable Label assistant (auto labling) Edit category Others1 Others2 Remove lable Select image by keyboard 2.2.2.1.1 Add category Click "add category" to add category for project Enter name and color for category setting, click OK when finish enter. 14 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.1.2 Add image(s) Choose adding picture by "image" or by "folder". 2.2.2.1.3 Labeling Tap on the target image and select category. Choose ideal labeling methods from pen, polygon, ellipse, rectangle or line. When using polygon for labeling, dot the dots along the target object edge. Once finish the labeling, please double-click the finish position. When using ellipse, rectangle, or line for labeling, move the cursor and hold mouse click until finish. 15 © 2022 ASUS, Inc.
ASUS, Inc . 2.2.2.1.4 Modify label When complete labeling by polygon, ellipse, rectangle and line, the boundary of labeled zone or shape will remain in dash type. User can still fine-tune and modify the labeling. Once complete, move the cursor outside the labeling area and tap to finish the modification. 2.2.2.1.5 Remove label To erase any unnecessary or wrong portion of labeling. Please move the cursor to the labelled area, and select "Erase".
ASUS, Inc . 2.2.2.1.6 Select image(s) using the keyboard Press up arrow button for last image. Press Ctrl+A to select all images at once. Press Ctrl and use cursor to select multiple images. 17 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.1.7 Label assistant To use label assistant for single image on the screen, click label assistant icon in the tool bar on the top. To use label assistant for multiple images, please select multiple images in the image list and click label assistant icon on top of the image list column. Tip: 1. Previous labelling would be overwriten by new operation. 2. Object(s) by label assistant will be automatically categorized as current category shown in category list. 3.
ASUS, Inc . 2.2.2.1.9 Others1 To delete image(s) from the image list, select target image(s) and click delete icon at the top of image list column. Label Detail can show labeled pixels for each category. 2.2.2.1.10 Others2 Auto saving available on the top left corner of tool bar. Once enable, image(s) will be automatically saved without dialog box. Number of image(s) and number of image(s) being labelled shown on the top row of image list column.
ASUS, Inc . 2.2.2.2 Labeller - Object Detection Allow users to label the required message in the images for Object Detection process. Add category Add image(s) Labeling Modify lable Label assistant (auto labling) Edit category Others1 Others2 Remove lable Select image by keyboard 2.2.2.2.1 Add category Click "add category" to add category for project. Enter name and colors for category setting, click OK when finish enter. 20 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.2.2 Add image(s) Choose adding picture by "image" or by "folder". 2.2.2.2.3 Labeling Drag cursor and select rectangle shape for object labeling. 21 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.2.4 Modify label After picture labelled with the rectangular selection, there is dot line to identify the boundary. Move cursor to the corner of rectangular box to fine tune selected area. Once complete, tap cursor outside of selected area. 2.2.2.2.5 Remove label To remove any unnecessary or wrong annotation during the labeling process. Please move the cursor to the labelled area, left click on annotation, and right click to select "Delete Current Annotation".
ASUS, Inc . 2.2.2.2.6 Select image(s) by keyboard Press up arrow button for last image. Press Ctrl+A to select all images at once. Press Ctrl and use cursor to select multiple images. 23 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.2.7 Label assistant To use label assistant for single image on the screen, click label assistant icon in the tool bar on the top. To use label assistant for multiple images, please select multiple images in the image list and click label assistant icon on top of the image list column. Tips: 1. Previous label will be erased if label assistant a previously labelled image. 2. Object(s) by label assistant will be automatically categorized as current category shown in category list. 3.
ASUS, Inc . 2.2.2.2.8 Modify category Click proper icon at top of category list column to delete or modify category. 2.2.2.2.9 Others-1 To delete image(s) from the image list, select target image(s) and click delete icon at the top of image list column. Label Detail can show number of bounding box for each category. 25 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.2.10 Others-2 Auto saving available on the top left corner of tool bar. Once enable, image(s) will be automatically saved without dialog box. Number of image(s) and number of image(s) being labelled shown on the top row of image list column. Image shown with green dot meaning label being saved. Bottom of the screen shows image information, eg image size, depth, and file size. 26 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.3 Labeller - Classification Allow users to label the required message in the images for Classification process. Add category Add image(s) Labeling Modify lable Label assistant Edit category others1 others2 Remove lable Select image by keyboard 2.2.2.3.1 Add category Click "Add category" to add category for project Enter name and colors for category setting, click OK when finish enter. 27 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.3.2 Add image(s) Select image category, choose add to images or files, import images to selected category. Images only shown in list when specific category chosen. 2.2.2.3.3 Labeling This is not a mandatory step. Drag cursor and use crop selection tool for labeling. Training will proceed entire photo without any specific object labelled. Tips: One object in one photo during classification process. Previous selection will be overwritten if 2nd selection proceed.
ASUS, Inc . 2.2.2.3.4 Modify labeling After picture labelled with the crop selection, there is dot line to identify the boundary. Move cursor to the corner of rectangular box to fine tune selected area. Once complete, tap cursor outside of selected area. 2.2.2.3.5 Remove annotation To remove any unnecessary annotation during the labeling process. Please move the cursor to the annotation, and right click to select "Clear All Annotation". 29 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.3.6 Select image(s) by buttons Up arrow button for last image; Down (right) arrow button for next image. Press Ctrl+A to select all image at once. Press ctrl and use cursor to select multiple images. 2.2.2.3.7 Label assistant To use label assistant for single image on the screen, click label assistant icon in the tool bar on the top.
ASUS, Inc . 2.2.2.3.8 Modify category Click proper icon at top of category list coulumn to delete or modify category. 2.2.2.3.9 Others-1 To delete image(s) from the image list, select target image(s) and click delete icon at the top of image list column. Label Detail can show number of images for each category. 31 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.3.10 Others-2 Auto saving available on the top left corner of tool bar. Once enable, image(s) will be automatically saved without dialog box. Number of image(s) and number of image(s) being labelled shown on the top row of image list column. Image shown with green dot meaning label being saved. Bottom of the screen shows image information, e.g. image size, depth, and file size. 32 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.4 Labeller - Anomaly Detection Allow users to label the required message in the images for Anomaly Detection process. Add image(s) Select image(s) Edit category others1 others2 2.2.2.4.1 Add image(s) Select image category, choose add to images or files, import images to selected category. Images only shown in list when specific category chosen. 33 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.4.2 Select image Up arrow button for last image; Down (right) arrow button for next image. Press Ctrl+A to select all images at once. Press Ctrl and use cursor to select multiple images. 2.2.2.4.3 Modify category Click proper icon at top of category list column to modify category color. 34 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.2.4.4 Others-1 To delete image(s) from the image list, select target image(s) and click delete icon at the top of image list column. Label Detail can show number of images for each category. 2.2.2.4.5 Others-2 Auto saving available on the top left corner of tool bar. Once enable, image(s) will be automatically saved without dialog box. Number of image(s) and number of image(s) being labelled shown on the top row of image list column. 35 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.3 Trainer This session is to train properly labeled image data prepared in labeler session for specific purpose AI model. You can prepare or train different AI model(s) by setting condition via different Task(s). Select task Create and delete task(s) Modify task detail Edit task(s) Start training Training process Task confirmation Training complete To scheduler 2.2.3.1 Select Task Define a task and select one as training boundary condition.
ASUS, Inc . 2.2.3.2 Create and delete task(s) Click "Add New Task(s)" (red box) to create a new task; Click "Remove Selected Task" (blue box) to delete current selected task. 2.2.3.3 How to modify task detail Click "Setting" button to see and edit the detail for the selected task. 37 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.3.4 Task detail confirmation Selected task name and detail in the center column in task setting (blue box below). Double check the data for training and verification in the right hand column (red box). Click any image file and preview it in the box below. 38 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.3.5 Edit task detail Turn on or off to define any parameter required for training. If trainer wish to manually adjust the data or train/verification ratio, please turn OFF training data ratio (blue arrow shown below). Click the right/left arrows at the top of the data list column to move the image file(s) desired. Click "Apply" icon for confirmation and "Back" icon to return to main page. 39 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.3.6 Start training Select training task and click "Start" button to start training. 2.2.3.7 Training process During training, please avoid any GPU hungry application(s). Top right corner shows the preprocess, training and calculating progress in percentage alongside a loss value curve shown in the center of training main page. For any expected training, the ideal curve will be trending down and a converging amplitude.
ASUS, Inc . 2.2.3.8 Training complete Once training progress completed, the percentage shows 100%. Ready for next step for verification. 2.2.3.9 Training schedule This step can schedule or unschedule multiple tasks in the Schedule. 41 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.4 Verifier This session is to verify how effective the AI model trained in Trainer session. Verification reference is the labeled "Ground Truth" in labeller session. Result pages show verifiction in Precision/Recall and Accuracy respectively. Select task Check the result of each image data Adjust overall threshold Apply default threshold value How to view report Detail report (1) Detail report (2) Export report Export model Confusion Matrix 2.2.4.
ASUS, Inc . 2.2.4.2 Check the result of each image data By clicking each image data, you can check the prediction result as verification purpose. 2.2.4.3 Adjust overall threshold Click verification "setting" icon to adjust threshold. Hit apply once threshold setting done. Screen shows current image and result accordingly. 43 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.4.4 Apply default threshold value Hitting "Default" can restore threshold and apply default value. 2.2.4.5 How to view report Click "View" icon to get into report page. It shows detail verification report. 44 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.4.6 Detail report (1) - (segmentation/Object detection) Report page shows Precision (Avg.) and Recall (Avg.) on top of this page and summary numbers right below. Right hand side shows detail of each image data for verification after prediction proceed upon category (Select desired category to see different result accordingly.) Row background color will display accurate prediction as green and inaccurate prediction in red accordingly upon desired or selected category.
ASUS, Inc . 2.2.4.7 Detail report (2) - (Classification and Anomaly detection) Report page shows Accuracy (Avg.) on top of this page and summary numbers right below. Right hand side shows ground truth vs prediction of each image data for verification after prediction proceed. (Select desired category to see different result accordingly.) Row background color will display accurate prediction as green and inaccurate prediction in red accordingly upon desired or selected category . 46 © 2022 ASUS, Inc.
ASUS, Inc . 2.2.4.8 Export report By clicking "Export Report" button and selecting target path, you can export report to a html format file. 2.2.4.9 Export model By clicking "Export" button and selecting target path, you can export current task model after a few minutes waiting. This model file can be used in Predictor or further use. 47 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.4.10 Export model for OpenVINO structure by AISVinoTool For AISVision v1.4 (launch in 2022/06), the inference site support the structure of OpenVINO, and please go to ASUS IoT website for AISVinoTools for the detail support material include manual and API reference. (https://iot.asus.com/products/AI-software/AISVision/) 2.2.4.11 Confusion Matrix Refer to the figure below for the definition and explanation of Confusion Matrix.
ASUS, Inc . False Negatives (FN): Predicted target event a Negative, and the actual event is a positive. 49 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . As mentioned above, we can use TP, TN, FP and FN values to calculate accuracy, recall, loss, detection and overkill: 1. Accuracy = (TP+TN) / (TP+FP+FN+TN) Accuracy is the ratio of total sum of all true events predicted over total sum of all predicted events. 2. Precision = TP / (TP+FP) Precision is the ratio of true positive events predicted over total sum of all predicted positive events. 3.
ASUS, Inc . Hence, the result as below 1. 2. 3. 4. 5. Accuracy = (TP+TN) / (TP+TN+FP+FN) = (5+90) / (5+90+3+2) = 95%, meaning 95% of total examination events are accurate result. Precision = TP / (TP+FP) = 5 / (5+3) = 62.5%, meaning 62.5% of total examined defect events are actually defect events. Recall or Detection = TP/ (TP+FN) = 5 / (5+2) = 71.4%, meaning 71.4% total actual defect parts examined as defect events. Loss = FN/ (TP+FN) = 2 / (5+2) = 28.6%, meaning 28.
ASUS, Inc . 2.2.5 Predictor Predictor allows you to proceed with the AI model (*.ditox generated by Verifier) for prediction to target data (image). The prediction result can be shown as an image or in text format. Choose model Select target data Proceed prediction Result check Adjust overall threshold Adjust default threshold value 2.2.5.1 Choose model (*.ditox) Click load model (red box) to select pre-trained model in .ditox sub-file name.
ASUS, Inc . 2.2.5.2 Select target data Click "Add image" icon(red box below) to select target image data you want to proceed prediction. Image data will be shown in below blue box once successfully load 2.2.5.3 Proceed prediction Click "Run" (below in red box) to proceed prediction. The time-lapse is depend on the model and data that had been set for prediction. Green marks are shown in front of predicted data image file(s) while red marks are for unpredicted data.
ASUS, Inc . 2.2.5.4 Result check Click the image(s) to check the prediction result. Result related information provided (in blue box below), and how much time spent shown in bottom. (yellow box) 2.2.5.5 Adjust overall threshold Click "Threshold Setting" icon to adjust threshold. Hit apply once threshold setting done. System will change all the images to unpredicted state, then return to step 3 to proceed prediction. 54 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.5.6 Adjust default threshold value Hitting "Default" can restore threshold and apply default value, then go back to step 3 to proceed prediction. 55 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.6 Scheduler This page allows user to manage multiple parameters upon pre-set project(s). Load project(s) Scheduling Re-arrange parameter(s) Training started 2.2.6.1 Load project(s) Click "load project" icon from project list panel, select project(s)(.ditprj) for training purpose. Tips: 1. Please check the project labeling for training purpose with proper parameters. If labeling not properly done yet, please circle back to Trainer 2. 3. function and re-do again.
ASUS, Inc . 2.2.6.2 Scheduling Select required parameter(s) to the scheduler list according to project(s) Available project(s) and parameter(s) shown in left hand side panel, select required parameter(s) according to project(s) selected. 2.2.6.3 Re-arrange parameter(s) Use the three (3) buttons on the top to move parameter(s) up and down, or delete parameter(s). Use "Cut-in" icon to intercept specific parameter(s) straight into the training sequence. 57 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . 2.2.6.4 Start training Once training parameter(s) arranged as plan, hit "START" button to start training. Show training progress in percentage on the side. Click right bottom corner "show result" button, tool can show training result after training finish. 58 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . Chapter 3: Upgrading AISVision 3.1 Complimentary AISVision upgrade policy You are eligible for a complimentary upgrade under the following circumstances. Otherwise, you will need to pay for an upgrade. A new version of AISVision is released with bug fixes A new version of AISVision is released with new features/functions/pre-trained models within 365 days of your AISVision purchase Otherwise, you will need to pay for an upgrade. 3.
ASUS, Inc . Step 2. 3.3 Cost to upgrade AI Toolkit Check with your ASUS representative for more details. 60 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . Chapter 4: Support for AISVision 4.1 Before you call customer support 4.1.1 Troubleshooting Double-check your hardware and software settings to make sure that they are set to run as designed. 4.1.2 Call for support Find the phone number of the ASUS IoT support center that is nearest to or most convenient for you in Contacting ASUS. A log file will be generated and saved in the system file manager folder (AISVision_version \App_Data\Logs). 4.1.
ASUS, Inc . 4.2 VGA QVL (Qualified Vendors List) list AISVision support NVIDIA® GeForce® 6GB above. And ASUS has verified the VGA card for the best stability and compatibility when installed the AISVision with VGA QVL list.
ASUS, Inc . Appendix Safety Information Regulatory notice Contacting ASUS Taiwan Talk to us AIS_support@asus.com Call Us | Official Support | ASUS Global China 官方支持 | ASUS 中国 Germany Telefonische Produktunterstützung +49 (0) 2102 5789 557 Telefon Ö sterreich +43 (0) 1360 2775 461 USA Chat with us 1-888-678-3688 https://www.asus.com/us/support 63 © 2022 ASUS, Inc. All rights reserved.
ASUS, Inc . Error Messages Do you want to save the modifications? FlowController.cs(177) Do not show this dialog box the next time, and project will be saved automatically every hour. The project has been modified. Do you want to save the FlowController.cs(215) modifications? Do not show this dialog box the next time, and project will be saved automatically. FlowController.cs(505) Would you save current project? FlowController.cs(540) The project is not found.
ASUS, Inc . PrjManagerForm.cs(738) Project is not fouund! ParamController.cs(519) Cannot delete following directory. [folder name] ParamController.cs(537) Cannot delete following directory. [folder name] SettingParameterForm.cs(385) Auto Split will Discard All Custom Changes. Continue Anyway? SettingParameterForm.cs(1096) The trained model will be deleted permanently (if exists), and it is irreversible.Continue anyway? SettingParameterForm.
ASUS, Inc . Cannot find any support image files in selected path VerifierController.cs(1321) (.png, .jpg, .jpeg). Please select a different path. VerifierController.cs(1353) Model weight (ckpt) is not defined. DoramiAIDetector.cs(65) Fail to load Model DoramiAIDetector.cs(185) Fail to load Image ClassEditor.cs(71) Class Name Can not be Blank ClassEditor.cs(88) The Class Name Already Exist. ClassEditor.cs(97) The Color of Class Already Exist. ClassEditor.