ContainersNVIDIADeepSeek-R1NVIDIALlama-3.1-Nemotron-70B-InstructNVIDIAPyTorch
CollectionsNVIDIAOmniverse Kit (FB)NVIDIADeepStream SDKNVIDIAOmniverse Kit App Streaming
ModelsNVIDIAStyleGAN3 pretrained modelsNVIDIAPeopleNetNVIDIATrafficCamNet
ResourcesNVIDIARiva Skills Quick Start
Helm ChartsNVIDIAGPU OperatorNVIDIANVIDIA NIM Operator
NVIDIA
NVIDIA
Optical Character Recognition
Model
NVIDIA
NVIDIA
Optical Character Recognition

Model to recognise characters from a preceding OCDNet model.

This model is backed by NVIDIA's Plus Plus (++) Promise
to learn more about the quality of the datasets used to train this model.
FieldResponse
Intended Application(s) & Domain(s):OCR: This model is intended to be used in computer vision application for recognizing the optical characters in a scene.
OCD: This model is intended to be used for detecting text in images.
Model Type:OCR: This model is intended for developers who want to recognize/identify text from image data.
OCD: This model is intended for developers who want to detect optical characters from image data.
Intended Users:OCR: This model is intended for developers who want to customize optical character recognition models.
OCD: This model is intended to be used for detecting text in images.
Output:OCR: Sequence of characters
OCD: BBox or polygon coordinates for each detected text in the input image
Describe how the model works:OCR: The training algorithm minimizes the connectionist temporal classification (CTC) loss between a ground truth character sequence from the image and a predicted characters sequence. Characters are decoded from the sequence output by a best-path decoding method.
OCD: Based on DBNet, a network architecture for real-time scene text detection, this model aims to solve the problem of text localization and segmentation in natural images with complex backgrounds and various text shapes.
Technical Limitations:OCR: This model performs best on images representing its training set: Uber Text and TextOCR. Uber Text contains street view images. TextOCR has images with text in various scenes. Further fine-tuning might be required for domain-specific accuracy.
OCD: The NVIDIA OCDNet trainable model is trained on Uber Text, which contains street-view images only. Further fine-tuning might be required for domain-specific accuracy.
Verified to have met prescribed NVIDIA standards:OCR: Yes
OCD: Yes
Performance Metrics:OCR: Accuracy
OCD: hmean
Potential Risks:OCR: None Known
OCR: None Known
Licensing:https://www.nvidia.com/en-us/data-center/products/nvidia-ai-enterprise/eula/