Resources » FAQ
AI.dielmo is a software developed by DIELMO 3D, with the objective of facilitating the incorporation of object detection based on Artificial Intelligence (AI) to any workflow. This allows you to start using AI from the get go without wasting large amounts of time setting up complex external dependencies.
You can perform actions based on the detected objects, such as cropping images with different parameters, selecting the best image among several object that are the same type, counting the number of objects in images, performing an image quality check to verify that they all contain a certain object, blurring objects (such as faces or license plates) within a list of images or videos, etc.
The Free and Premium versions are demo versions with standard workflows, which can be easily customized for each client (Custom version).
AI.dielmo can be used as an executable with our graphical interface that allows the management of process queues or it can be used as a library to integrate object detection in other software packages, counting on our technical support.
The software installation is done through an executable installer, which can be downloaded here for free.
The installer includes all the dependencies needed for the artificial intelligence object detection to work and there is no need to install/configure complex external dependencies.
AI.dielmo has 3 types of licenses:
Object detection is a general term used to describe a set of computer vision techniques aimed at identifying objects of a certain class (such as humans, buildings or cars) in videos and digital images.
In principle, we can train a model to recognize any kind of objects in images as long as we are able to provide a large number of sample images where the object to be detected is clearly identified.
The process of training a model needs a large number of different images and their labelings where the object is in the image. Currently there are many models that come pre-trained to detect different types of objects, but we can perform customized training to detect anything you need.
To start detecting objects in your images, you will need a pre-trained model, you can find a list of our models here.
Once you have your model, simply select the type of process you want to perform, select the model you have downloaded (depending on the model it will detect some objects or others), select the input data, and an output folder. Then run the task.
By default there are different types of tasks, such as cropping the best object in the image, drawing the detections on the input image or video, blurring the objects on the input images or videos, etc.
While a calculation is being performed, new tasks can be queued to be executed when the running process is finished.
Also, with the custom version, we can tailor new tasks to fit the needs of each customer.
If you perform repetitive tasks where a person has to manually go over several images and analyze them, this software will save you time and money.
It can be trained to perform any task you need, such as filtering every picture that has a certain object in them, quality control tasks or even having the option to blurry faces or license plates automatically.
You can also use our software to perform different proofs of concept of the use of artificial intelligence based on object detection and see how it could be integrated into your workflow. We can help you adapt our software to your needs or collaborate in your research and development projects.
AI.dielmo has been designed to satisfy all technical levels, from the most inexperienced to the most advanced users.
Each detection can be specifically configured under certain parameters and specifications. In addition, there are several detections such as, for example, clipping the best object in the image, drawing the detections on the input image or video, blurring the objects on the input images or videos, etc.
Thanks to the custom version, we can customize new tasks to fit the needs of each customer.
A model consists of a series of mathematical operations that are applied on the input data to predict an output.
In principle, we can train a model to recognize any type of objects in images, as long as we are able to provide a large number of sample images where the object to be detected is clearly identified.
The process of training a model consists of providing a large number of different images and labeling the location where the object is in the image.
Currently there are a large number of models already pre-trained to detect different types of objects, but we can perform customized training to detect specific objects.
To simplify the use of the software and to be sure that all the models work properly, AI.dielmo uses a proprietary format (.dmod) to store the models and it is not possible to use third party models.
But if you have trained your own or third party object detection models based in tensorflow, we can convert them to our format and you will be able to use them in AI.dielmo. We also can support you to do your own model trainings or train them for you.
When you are going to launch one of the different calculations available in AI.dielmo, the first parameter you will need to indicate will be the input path, which can be an image folder or a video. You can also directly drag and drop a folder or file.
All the images that are inside the selected folder and its subfolders will be analyzed.
The calculation speed depends on many factors such as the type of model or process to be performed, the processor, the hard disk speed, whether the calculation is performed locally or in a network folder, the amount of RAM available, etc.
To achieve the highest speed in image analysis, it is convenient to have a graphics card (GPU) that allows working with CUDA, including the NVIDIA CUDA Deep Neural Network library (cuDNN). “Here you can find more information on how to configure the GPU.
The software is intended to be easily modifiable in any of the workflow phases (way of iterating, input formats, pre-processing of images before object detection, training of customized models, post-processing of images after object detection, output format, generation of customized reports, etc).
Currently we have dozens of models completely free of charge. You can see the list of models here.
We can also customize models according to your needs or teach you how to train your own models.
Yes, both the software and dozens of models are available for free download and without registration.
When launching a calculation you must specify the output path where the results will be saved.
Depending on the type of process to be executed, you will obtain different types of outputs, which can be images, videos, vector files, etc. In addition, you will always find a report in .csv format together with the outputs, which includes the information relevant to the task performed.
Yes, in the premium version subscription, you can get a discount equivalent to two months if you pay in advance on an annual basis.
There are different types of advanced parameters depending on the task to be executed. If you hover the mouse over each of them you will see a description of the parameter. For more information, you can consult the manual or the video tutorials at our resources page.
Whenever you launch a new task, it is pre-configured with default parameters. If you click on the advanced parameters option you can modify them. If you want to launch a new job with the same parameters as a previous job, you can use the clone option. For more information, you can consult the manual or the video tutorials at our resources page.
For proof-of-concept testing, you can use any computer. In a production environment you must take into account that the calculation speed depends on many factors such as the type of model or process to be performed, the processor, the hard disk speed, whether the calculation is performed locally or in a network folder, the amount of RAM available, etc.
To achieve the highest speed in image analysis it is convenient to have a graphics card (GPU) that allows working with CUDA, including the NVIDIA CUDA Deep Neural Network library (cuDNN). “Here you can find more information on how to configure the GPU.
Whenever you launch a new task, it is pre-configured with default parameters. If you click on the advanced parameters option, you can modify them. If you want to launch a new job with the same parameters as a previous job, you can use the clone option.
There are too many variables to define which one works best with a particular type of object, so we recommend that you consult the manual or the video tutorials “Here”.
The detection rate will depend on each model. Many of them are above 99%.
Yes, you can install AI.dielmo on any of the three mentioned operating systems.
To simplify the use of the software and to be sure that all the models work properly, AI.dielmo uses a proprietary format (.dmod) to store the models and it is not possible to use third party models.
But if you have trained your own or third party object detection models based in tensorflow, we can convert them to our format and you will be able to use them in AI.dielmo. We also can support you to do your own model trainings or train them for you.
The Free and Premium versions are demo versions with standard workflows, which can be easily customized for each client (Custom version) by changing any aspect of the workflow before and after the detections.
Currently there is a large number of pre-trained models for detecting different types of objects, but we can perform custom training for detecting specific objects or help you to train new models yourself.
Currently there is a large number of pre-trained models for detecting different types of objects, but we can perform custom training for detecting specific objects or help you to train new models yourself.
If you need to report a bug or have any suggestions, you can send us a ticket through your accounts’ ticketing system.
AI.dielmo can be used for an infinite number of applications. It is currently being used professionally in the cropping of images in the environment of an electrical tower, selection of the best photograph among several of the same object, blurring of faces or license plates in images or videos, etc.
The free version is limited to 500 detections per day. This will allow you to enter the world of AI-based object detection and test if the software fits your needs. If you need unlimited detections, we offer a subscription to the premium service for a very affordable price.
Surely when testing our software with your images you will come up with new functionalities that will help you to improve your production processes, for this reason we offer you the “custom” version that will adapt to your specific needs.
Confidence measures the accuracy of the detections made by the AI model. Range [0-100%].
A confidence value close to 100% indicates that all detections made will most certainly be of the type of object we are looking for. A lower confidence value indicates that there is a margin of doubt that it may be that type of object or not.
Lowering this threshold will result in more detections, but it is also possible that some of them will be mistaken for another type of object. For this reason, depending on the model you are using and your needs, you can set the appropriate confidence value.
Only detections with a precision higher than the confidence threshold will be used for the rest of the processes. Those that do not exceed the threshold will be discarded.
We use cookies on our website to give you the most relevant experience by remembering your preferences and repeat visits.
By clicking “Accept”, you consent to our cookies policy.