![[Neurocle] Logo_white.png](https://static.wixstatic.com/media/242158_0a4304ae861e4107925bc861d88c280f~mv2.png/v1/fill/w_150,h_21,al_c,q_85,usm_0.66_1.00_0.01,enc_auto/%5BNeurocle%5D%20Logo_white.png){kind=small}
New Features of v4.1
Ready for a new experience?
Divide high-resolution images
into multiple patches for training
Patch Classification Model
Patch Classification Model is used when high-resolution images contain tiny defects.
This technique minimizes missed defects and helps to identify and distinguish between normal and defective areas within images.
Learn defects without any loss of areas
Patch Mode of GAN Model
When training the GAN model, the image is divided into patches to learn defects. This method ensures that defects are learned without any loss of areas, enabling the generation of more accurate virtual defect images and improving model accuracy.
Support for various operating systems
and platforms
Runtime Processor Expansion
Supports Runtime processors such as NVIDIA CUDA, OpenVINO, and DirectML. Additionally, from v4.1, we expanded platform support from CPUs, GPUs to embedded boards and NPUs, providing more options for model inference processors.
.png)
Convert the shape of a label
with a single click
Shape Converter
Convert box labels into shapes that fit the objects with a single click. Shape converter ensures accurate labeling in the shortest time possible and significantly reduces manual effort.
Simplify model operations to reduce
size and weight
Quantization
Quantization can be used when fast inference speed is required. Quantization lowers computational load, while maintaining high accuracy and speed. Thus, it makes models lightweight and suitable for diverse applications.
