NE101 vs NE301 vs NE302: Which CamThink Edge Vision Camera Should You Choose?

Choose by the Job the Camera Will Do

Choose NeoEyes NE101 when your system needs low-frequency images for analysis elsewhere. Choose NeoEyes NE301 when the camera must run AI locally and the installation calls for an outdoor enclosure, battery operation, LTE, or PoE. Choose NeoEyes NE302 when you need local AI in a compact indoor device or terminal.

The closest comparison is NE301 versus NE302: both use an STM32N6 with Neural-ART acceleration. Their different memory, enclosure, power, and expansion designs matter more to product selection than their shared 0.6 TOPS figure. A higher model number does not describe a universal upgrade path.

Start with the output your application needs, then consider where the camera will live. A meter image sent to an existing OCR service, an outdoor camera reporting a detection, and a small terminal recognizing an equipment state place different demands on the hardware.

NE101, NE301, and NE302 at a Glance

The table compares the documented standard camera platforms. Optional communication boards and power accessories are identified separately from the baseline hardware.

Specifications: NE101 Product Information, NE301 Product Information, and NE302 Product Information. NE302 radio and regional configuration follow the selected SKU. Enclosure dimensions describe the housing; installation also needs space for antennas, cables, and mounting.

Sensor resolution needs context, too. NE101’s 5 MP sensor can capture a detailed still image, while NE301 and NE302 add NPU-assisted processing to their 4 MP image paths. Target detail depends on the lens, distance, lighting, and pixels occupied by the object. The image fed to an AI model may be resized, so megapixels alone cannot predict recognition accuracy.

When NE101 Is the Right Starting Point

NE101 is a strong fit when image acquisition is the camera’s main job. It can wake on a schedule or supported trigger, capture an image, send it over the configured network, and return to sleep. Analysis can run on your existing server, gateway, or cloud application.

Consider a meter-reading project that already has an OCR service. The camera needs to deliver a readable image at the required interval. Putting another model inside the camera may add deployment work without changing the result the business receives. NE101 lets the team concentrate on focus, illumination, capture timing, and image delivery.

NE101 also has an ESP-DL face-detection development example. Describing it as incapable of any local AI would be inaccurate. Its ESP32-S3 development path is different from the dedicated Neural-ART acceleration and model-deployment workflows offered by NE301 and NE302.

If the camera must produce a detection result before transmitting, evaluate an STM32N6 model on NE301 or NE302. Sending a structured result can reduce the need to transmit full images, while selected images can still provide evidence for review.

NE301 vs NE302: Same Accelerator, Different Hardware Priorities

NE301 for a camera deployed at the site

NE301 packages local inference in an IP67 camera platform with a battery tray, mounting accessories, and optional LTE Cat.1 or PoE connectivity. Its modular design also supports camera and sensor expansion. These features are useful when the camera needs its own mounting point, power arrangement, and network connection.

An outdoor equipment-monitoring point is a typical example. The model matters, but so do the sealed enclosure, physical installation, and a way to get results off site. NE301 provides a documented starting point for those deployment decisions.

NE302 for vision inside an indoor product

NE302 concentrates the camera, STM32N6, STM32U0 control MCU, memory, and wireless circuits on a 38 × 38 mm Main Board. The complete device includes an Interface Board with USB Type-C power, MicroSD, and development connections, inside a 42 × 42 × 20 mm enclosure.

That footprint is useful for a self-service terminal or workstation with an existing power supply and a defined camera position. Teams can evaluate the complete device first, then assess Main-Board-only integration with DC power. MicroSD access belongs to the Interface Board, so a board-only design needs an appropriate storage plan.

The standard camera can serve as evaluation hardware; CamThink can also help assess an application-specific hardware and integration configuration when the standard setup does not fit.

Memory still matters when TOPS are equal

NE301 has 64 MB PSRAM and 128 MB Flash, compared with NE302’s 32 MB PSRAM and 64 MB Flash. That gives NE301 more external memory and storage headroom for application code, model assets, and buffers. It does not establish a fixed performance advantage for every model.

Both cameras require models prepared for their supported runtime and memory layout. Equal accelerator ratings do not establish interchangeable firmware or model binaries. For NE302, a raw ONNX or TFLite file must be converted and packaged for the device before upload. Choose using the actual model’s memory use, supported operators, preprocessing, and result path, rather than inferring frame rate from TOPS.

For a closer look at the compact platform, use the NE302 Quick Guide to try camera preview and model validation.

Let Power and Connectivity Narrow the Choice

For a remote point without continuous power, NE101 and NE301 offer documented AA-battery deployment paths. Choose between them according to whether inference belongs at the camera. Battery runtime still depends on wake frequency, active time, lighting, radio conditions, and uploads; a sleep-current figure alone cannot describe the system’s operating life.

For a powered indoor terminal, NE302 can use continuous USB Type-C power or an external USB-C battery pack. Its STM32U0 supports sleep and wake control, but the product is not supplied as an outdoor AA-battery camera. The host device’s power and space constraints usually determine whether the complete two-board assembly or custom Main Board integration makes more sense.

If LTE or PoE is a firm requirement, NE301 provides the documented local-AI route among these three products. NE101 also offers LTE Cat.1 for image delivery. NE302 has Wi-Fi and separate Wi-Fi HaLow variants; HaLow needs matching access-point infrastructure and the appropriate regional radio configuration. It is not a cellular connection.

A live preview or video stream can help commissioning without making a camera a complete continuous-video analytics system. If your requirement depends on uninterrupted motion history, tracking, or multiple video workloads, first revisit the architecture in Event-Triggered vs Continuous Video Monitoring.

Which Camera Fits Your Project?

The following examples illustrate selection logic. They describe starting points for evaluation, not preconfigured applications or customer deployment results.

The same application name can lead to different hardware. A meter project can use NE101 when OCR runs on a server, NE301 when recognition must run at an outdoor camera, or NE302 when the camera becomes part of an indoor instrument. Decide where the work happens before choosing the product.

Start with One Image and the Output You Need

Begin with an image taken from the intended mounting position and a clear output requirement: a readable photograph, a detection result, or a reported equipment state. For NE101, the first useful proof is that the receiving application gets the required image. For NE301 and NE302, add a compatible model and confirm that the application receives its result.

Use that working path to evaluate the chosen power and network configuration. It gives the team something concrete to improve: image quality, inference behavior, installation fit, or delivery timing. CamThink’s product documentation, model resources, and evaluation support provide the next steps for the selected camera.