Use the NGC CLI to download:
Copied!
Model
Resnet18 model to classify a car crop into 1 out 20 car brands.
Use the NGC CLI to download:
Copied!
5 Versions
pruned_onnx_v1.1.0Signed
This image has a digital signature verifying that it has not been altered or corrupted since its signing.
09/20/2024 1:06 AM UTC7.07 MB
Copied!
11/02/2023 10:27 PM UTC16.45 MBAccuracy: 9180 Epochs
Copied!
10/26/2021 4:03 PM UTC16.45 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100Copied!
10/26/2021 4:03 PM UTC16.45 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100
Copied!
1. MODEL OVERVIEW
| Key | Value |
|---|---|
| 1. Description | CLASSIFIES CARS BY MAKE |
| 2. Model Architecture | CONVOLUTION NEURAL NETWORK (CNN); Classification model with Resnet18 backbone |
| 3. Input | TYPE: RED, GREEN, BLUE (RGB) IMAGE; DIMENSION: 2D; SENSOR: RGB CAMERA; RESOLUTION(S): 224X224X3 |
| 4. Output | TYPE: LABEL; DIMENSION: 2D; RESOLUTION(S): 1X20 |
| 5. RUNTIME(S) | DEEPSTREAM |
| 6. Supported OS | LINUX, LINUX 4 TEGRA |
| 7. Supported Hardware | ALL NVIDIA GPUS INCLUDING JETSON DEVICES |
2. TRAINING & EVALUATION
| Key | Value |
|---|---|
| 1. TRAINING FRAMEWORK | KERAS AND TENSORFLOW |
| 2. TRAINING ARITHMETIC PRECISION | FP32 |
| 3. TRAINING DATASET | PROPRIETARY |
| 4. TRAINING DATASET PROPERTIES | 60,000 IMAGES; IMAGES OF VEHICLES; CONTAINS CAR CROPS FROM A MIX OF CAMERA HEIGHTS, CAMERA ANGLES, FIELD-OF VIEW (FOV) AND OCCLUSIONS. |
| 5. TRAINING DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
| 6. TRAINING KPI(S) | CATEGORICAL CROSS ENTROPY |
| 7. EVALUATION DATASET | PROPRIETARY |
| 8. EVALUATION DATASET PROPERTIES | 2,000 IMAGES ACROSS A VARIETY OF ENVIRONMENTS, OCCLUSION CONDITIONS, CAMERA HEIGHTS AND CAMERA ANGLES. |
| 9. EVALUATION DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
3. INFERENCE/PERFORMANCE
| Key | Value |
|---|---|
| 1. ENGINE(S) | TENSORRT (TRT) |
| 2. TEST HARDWARE | AGX XAVIER, NANO, T4, XAVIER NX |
| 3. ARITHMETIC PRECISION | AGX XAVIER: INT8, NANO: FP16, T4: INT8, XAVIER NX: INT8 |
| 4. INFERENCE PERFORMANCE | Frame Rate (Batch size =1): AGX XAVIER: 5746 FPS, NANO: 120 FPS, T4: 15743 FPS, XAVIER NX: 2689 FPS |
| 5. KPI(S) | ACCURACY: 91 |
pruned_v1.0Selected08/18/2021 8:10 PM UTC16.45 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100Copied!
pruned_v1.0Selected
08/18/2021 8:10 PM UTC16.45 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100
Copied!
1. MODEL OVERVIEW
| Key | Value |
|---|---|
| 1. Description | CLASSIFIES CARS BY MAKE |
| 2. Model Architecture | CONVOLUTION NEURAL NETWORK (CNN); Classification model with Resnet18 backbone |
| 3. Input | TYPE: RED, GREEN, BLUE (RGB) IMAGE; DIMENSION: 2D; SENSOR: RGB CAMERA; RESOLUTION(S): 224X224X3 |
| 4. Output | TYPE: LABEL; DIMENSION: 2D; RESOLUTION(S): 1X20 |
| 5. RUNTIME(S) | DEEPSTREAM |
| 6. Supported OS | LINUX, LINUX 4 TEGRA |
| 7. Supported Hardware | ALL NVIDIA GPUS INCLUDING JETSON DEVICES |
2. TRAINING & EVALUATION
| Key | Value |
|---|---|
| 1. TRAINING FRAMEWORK | KERAS AND TENSORFLOW |
| 2. TRAINING ARITHMETIC PRECISION | FP32 |
| 3. TRAINING DATASET | PROPRIETARY |
| 4. TRAINING DATASET PROPERTIES | 60,000 IMAGES; IMAGES OF VEHICLES; CONTAINS CAR CROPS FROM A MIX OF CAMERA HEIGHTS, CAMERA ANGLES, FIELD-OF VIEW (FOV) AND OCCLUSIONS. |
| 5. TRAINING DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
| 6. TRAINING KPI(S) | CATEGORICAL CROSS ENTROPY |
| 7. EVALUATION DATASET | PROPRIETARY |
| 8. EVALUATION DATASET PROPERTIES | 2,000 IMAGES ACROSS A VARIETY OF ENVIRONMENTS, OCCLUSION CONDITIONS, CAMERA HEIGHTS AND CAMERA ANGLES. |
| 9. EVALUATION DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
3. INFERENCE/PERFORMANCE
| Key | Value |
|---|---|
| 1. ENGINE(S) | TENSORRT (TRT) |
| 2. TEST HARDWARE | AGX XAVIER, NANO, T4, XAVIER NX |
| 3. ARITHMETIC PRECISION | AGX XAVIER: INT8, NANO: FP16, T4: INT8, XAVIER NX: INT8 |
| 4. INFERENCE PERFORMANCE | Frame Rate (Batch size =1): AGX XAVIER: 5746 FPS, NANO: 120 FPS, T4: 15743 FPS, XAVIER NX: 2689 FPS |
| 5. KPI(S) | ACCURACY: 91 |
08/18/2021 8:10 PM UTC132.46 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100Copied!
08/18/2021 8:10 PM UTC132.46 MBAccuracy: 9180 EpochsBatch Size: 1GPU: V100
Copied!
1. MODEL OVERVIEW
| Key | Value |
|---|---|
| 1. Description | CLASSIFIES CARS BY MAKE |
| 2. Model Architecture | CONVOLUTION NEURAL NETWORK (CNN); Classification model with Resnet18 backbone |
| 3. Input | TYPE: RED, GREEN, BLUE (RGB) IMAGE; DIMENSION: 2D; SENSOR: RGB CAMERA; RESOLUTION(S): 224X224X3 |
| 4. Output | TYPE: LABEL; DIMENSION: 2D; RESOLUTION(S): 1X20 |
| 5. RUNTIME(S) | DEEPSTREAM |
| 6. Supported OS | LINUX, LINUX 4 TEGRA |
| 7. Supported Hardware | ALL NVIDIA GPUS INCLUDING JETSON DEVICES |
2. TRAINING & EVALUATION
| Key | Value |
|---|---|
| 1. TRAINING FRAMEWORK | KERAS AND TENSORFLOW |
| 2. TRAINING ARITHMETIC PRECISION | FP32 |
| 3. TRAINING DATASET | PROPRIETARY |
| 4. TRAINING DATASET PROPERTIES | 60,000 IMAGES; IMAGES OF VEHICLES; CONTAINS CAR CROPS FROM A MIX OF CAMERA HEIGHTS, CAMERA ANGLES, FIELD-OF VIEW (FOV) AND OCCLUSIONS. |
| 5. TRAINING DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
| 6. TRAINING KPI(S) | CATEGORICAL CROSS ENTROPY |
| 7. EVALUATION DATASET | PROPRIETARY |
| 8. EVALUATION DATASET PROPERTIES | 2,000 IMAGES ACROSS A VARIETY OF ENVIRONMENTS, OCCLUSION CONDITIONS, CAMERA HEIGHTS AND CAMERA ANGLES. |
| 9. EVALUATION DATASET LICENSE | COVERED BY MODEL END USER LICENSE AGREEMENT (EULA) |
3. INFERENCE/PERFORMANCE
| Key | Value |
|---|---|
| 1. ENGINE(S) | TENSORRT (TRT) |
| 2. TEST HARDWARE | AGX XAVIER, NANO, T4, XAVIER NX |
| 3. ARITHMETIC PRECISION | AGX XAVIER: INT8, NANO: FP16, T4: INT8, XAVIER NX: INT8 |
| 4. INFERENCE PERFORMANCE | Frame Rate (Batch size =1): AGX XAVIER: 5746 FPS, NANO: 120 FPS, T4: 15743 FPS, XAVIER NX: 2689 FPS |
| 5. KPI(S) | ACCURACY: 91 |