Tech
Chameleon
Chameleon: Ultra-compact, Low-power, High-performance Edge Platform for On-Device Multimodal AI
Sensor
Chameleon provides robust sensor fusion capabilities, significantly enhancing the detection performance of AI systems.
Software
Chameleon delivers robust real-time computing through heterogeneous core architectures, optimized device porting, and advanced network slimming technologies.
Hardware
Our ultra-compact, ultra-low-power hardware with Embedded Linux and ROS middleware delivers optimized perception and efficient edge computing capabilities.
Tech
Chameleon
Chameleon: Ultra-compact, Low-power, High-performance Edge Platform
for On-Device Multimodal AI
Sensor
Chameleon provides robust sensor fusion capabilities, significantly enhancing the detection performance of AI systems.
Software
Chameleon delivers robust real-time computing through heterogeneous core architectures, optimized device porting, and advanced network slimming technologies.
Hardware
Our ultra-compact, ultra-low-power hardware with Embedded Linux and ROS middleware delivers optimized perception and efficient edge computing capabilities.
AI Solutions Designed to Fit Any Scenario
Main Sensor Options - RGB Camera, LWIR Camera, MIC, LiDAR, Ultrasonic Camera, TVOC Sensor
RGB+MICEquipment Failure Detection
RGB+LWIRFlood Detection
RGB+LWIR
Anomalous Heat Source Detection
RGB+초음파카메라
Pipe Leakage and Electrical Fault Detection
Optimal combinations of sensors provide the best solutions for a variety of situations.
AI Solutions Designed to Fit Any Scenario
Main Sensor Options - RGB Camera, LWIR Camera, MIC, LiDAR, Ultrasonic Camera, TVOC Sensor
RGB+MIC
Equipment Failure Detection
RGB+LWIR
Flood Detection
RGB+LWIR
Anomalous Heat Source Detection
RGB+초음파카메라
Pipe Leakage and Electrical
Fault Detection
Optimal combinations of sensors provide the best solutions
for a variety of situations.
Chameleon keeps evolving.
Spec
Chameleon 7 | Chameleon 7 Lite | Chameleon 8 |
Chameleon 7 (QCS605)
Released in 2021.
CPU
64bit 8 Core Qualcomm®Kryo™
300 - 10nm LPE
2 x Gold @ 2.5 GHz
6 x Silver @ 1.7 GHz
GPU
Qualcomm® Adreno™ GPU 615@780MHz
DSP
Hexagon 685 DSP with dual Hexagon Vector eXtensions (HVX)
Memory
LPDDR4x 4GB + eMMC 64GB
Connectivity
WLAN : Wi-Fi 4, Wi-Fi 5 (802.11ac) 2x2
BT : Bluetooth 5.1
Camera
Spectra270 Dual ISP; 2x 16MP, 1x 32MP
4K 60fps video (HEVC enc/dec)
Dimension
42mm x 35mm
OS Support
Android
Chameleon 7 Lite (QCS610)
Released in 2023.
CPU
64bit 8 Core Qualcomm® Kryo™
460 - 11nm LPP
2 x Gold @ 2.2GHz
6 x Silver @ 1.8GHz
GPU
Qualcomm® Adreno™ GPU 612@845MHz
DSP
Hexagon DSP with dual Hexagon Vector eXtensions (HVX), 1.1Ghz
Memory
LPDDR4x 4GB + eMMC 64GB
Connectivity
WLAN : Wi-Fi 4, Wi-Fi 5 (802.11ac) 1x1
BT : Bluetooth 5.0
Camera
Spectra250L Triple ISP; 16+16MP, 24MP@30fps
4K 30fps video (HEVC enc/dec)
Dimension
50mm x 38mm
OS Support
Linux Embedded
Chameleon 8 (QCS6490)
Scheduled for release in 2024.
CPU
64bit 8 Core Qualcomm® Kryo™
670 - 6nm
1 x Prime @ 2.7GHz
3 x Gold @ 2.4GHz
4 x Silver @ 1.9GHz
GPU
Qualcomm® Adreno™ GPU 643@812MHz
DSP
6th gen Qualcomm AI Engine
Compute Hexagon DSP with dual Hexagon Vector eXtensions (HVX), Hexagon Co-processor (Hexagon CP) 2.0, 4K Hexagon Matrix eXtensions (HMX), Hexagon Tensor Accelerator (HTA)
Memory
LPDDR4x 8GB + UFS2.2 128GB
Connectivity
WLAN : Wi-Fi 6 (802.11ax) and Wi-Fi 6E
BT : Bluetooth 5.2
Camera
Spectra570L Triple ISP; 3x22 MP@30fps,
36MP@30fps+22MP@30fps, 64MP@30fps
ZSL 4K 60fps decode / 4K 30fps encode
Dimension
TBD
OS Support
Linux Ubuntu
Tech
wNet
Advancing specialized Deep Neural Networks (DNNs) for targeted problem-solving.
Applying our custom neural model
wNet in diverse projects.
Bottom-center-based precise location
detection of road-moving objects
Detection of flood risk through river
width calculation.
Belt conveyor anomaly diagnosis through abnormal sound detection.
Anomalous heat source detection through
flame detection technology.
Proprietary large-scale dataset
from field-proven projects.
Tech
Network Slimming
Edge performance is preserved through model quantization and pruning.
Quantization
Post-training quantization applied for DSP deployment.
Pruning
Maintained detection accuracy at 90.2% AP@0.5 after pruning (down from 90.8%), while more than doubling inference speed.
Chameleon keeps evolving.
Spec
![]() | ![]() | ![]() | |
Chameleon 7 (QCS605) Released in 2021. | Chameleon 7 Lite (QCS610) Released in 2023. | Chameleon 8 (QCS6490) Scheduled for release in 2024. | |
CPU | 64bit 8 Core Qualcomm®Kryo™ 300 - 10nm LPE 2 x Gold @ 2.5 GHz 6 x Silver @ 1.7 GHz | 64bit 8 Core Qualcomm® Kryo™ 460 - 11nm LPP 2 x Gold @ 2.2GHz 6 x Silver @ 1.8GHz | 64bit 8 Core Qualcomm® Kryo™ 670 - 6nm 1 x Prime @ 2.7GHz 3 x Gold @ 2.4GHz 4 x Silver @ 1.9GHz |
GPU | Qualcomm® Adreno™ GPU 615@780MHz | Qualcomm® Adreno™ GPU 612@845MHz | Qualcomm® Adreno™ GPU 643@812MHz |
DSP | Hexagon 685 DSP with dual Hexagon Vector eXtensions (HVX) | Hexagon DSP with dual Hexagon Vector eXtensions (HVX), 1.1Ghz | 6th gen Qualcomm AI Engine Compute Hexagon DSP with dual Hexagon Vector eXtensions (HVX), Hexagon Co-processor (Hexagon CP) 2.0, 4K Hexagon Matrix eXtensions (HMX), Hexagon Tensor Accelerator (HTA) |
Memory | LPDDR4x 4GB + eMMC 64GB | LPDDR4x 4GB + eMMC 64GB | LPDDR4x 8GB + UFS2.2 128GB |
Connectivity | WLAN : Wi-Fi 4, Wi-Fi 5 (802.11ac) 2x2 BT : Bluetooth 5.1 | WLAN : Wi-Fi 4, Wi-Fi 5 (802.11ac) 1x1 BT : Bluetooth 5.0 | WLAN : Wi-Fi 6 (802.11ax) and Wi-Fi 6E BT : Bluetooth 5.2 |
Camera | Spectra270 Dual ISP; 2x 16MP, 1x 32MP 4K 60fps video (HEVC enc/dec) | Spectra250L Triple ISP; 16+16MP, 24MP@30fps 4K 30fps video (HEVC enc/dec) | Spectra570L Triple ISP; 3x22 MP@30fps, 36MP@30fps+22MP@30fps, 64MP@30fps ZSL 4K 60fps decode / 4K 30fps encode |
Dimension | 42mm x 35mm | 50mm x 38mm | 45mm x 45mm |
OS Support | Android | Linux Embedded | Linux Ubuntu |
Tech
wNet
Advancing specialized Deep Neural Networks (DNNs) for targeted problem-solving.
Applying our custom neural model wNet in diverse projects.
Bottom-center-based precise location detection of road-moving objects
Detection of flood risk through
river width calculation.
Belt conveyor anomaly diagnosis through abnormal sound detection.
Anomalous heat source detection through flame detection technology.
Proprietary large-scale dataset from field-proven projects.
Tech
Network Slimming
Edge performance is preserved through model quantization and pruning.
Quantization
Post-training quantization applied for DSP deployment.
Pruning
Maintained detection accuracy at 90.2% AP@0.5 after pruning (down from 90.8%), while more than doubling inference speed.
Headquartersㅣ4F, BK Tower, 28 Beopwon-ro 11-gil, Songpa-gu, Seoul (05836) Pohang Branch Officeㅣ#303, Venture Building 5, Pohang Technopark, 394 Jigok-ro, Nam-gu, Pohang, Gyeongsangbuk-do, Seoul |
© 2024 WITHROBOT. All Rights Reserved.