Hugging Face's logo

STMicroelectronics
/
deeplab_v3

Image Segmentation
Model card Files Community
FBAGSTM commited on
Commit
955ec6c
·
verified ·
1 Parent(s): dea1725

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +23 -23
README.md CHANGED
@@ -70,9 +70,9 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
70
 
71
  | Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STM32Cube.AI version | STEdgeAI Core version |
72
  |------------|---------------|----------|------------|-----------|--------------|--------------|---------------|----------------------|-----------------------|
73
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 256x256x3 | STM32N6 | 2253.5 | 0.0 | 1001.25 | 10.0.0 | 2.0.0 |
74
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) |person COCO 2017 + PASCAL VOC 2012 | Int8 | 320x320x3 | STM32N6 | 2446.0 | 0.0 | 1000.41 | 10.0.0 | 2.0.0 |
75
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 416x416x3 | STM32N6 | 2743.5 | 2028.0 | 2721.19 | 10.0.0 | 2.0.0 |
76
 
77
 
78
 
@@ -81,24 +81,24 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
81
 
82
  | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
83
  |------------|---------------|----------|------------|------------------|------------------|---------------------|-------------|----------------------|-------------------------|
84
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 27.36 | 36.54 | 10.0.0 | 2.0.0 |
85
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 44.99 | 22.22 | 10.0.0 | 2.0.0 |
86
- | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 191.91 | 5.21 | 10.0.0 | 2.0.0 |
87
 
88
 
89
  ### Reference **MPU** inference time based on COCO 2017 + PASCAL VOC 2012 segmentation dataset 21 classes and a derivative person dataset from it (see Accuracy for details on dataset)
90
  | Model | Dataset | Format | Resolution | Quantization | Board| Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version |Framework |
91
  |----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------|--------|------------|----------------|-------------------|------------------|-----------|---------------------|-------|--------|------|--------------------|-----------------------|
92
  | [DeepLabV3 per tensor (no ASPP)](https://www.st.com/en/embedded-software/x-linux-ai.html) | COCO 2017 + PASCAL VOC 2012 | Int8 | 257x257x3 | per-tensor | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 52.75 | 99.2 | 0.80 | 0 | v5.1.0 | OpenVX | | | | | v5.1.0
93
- | [DeepLabV3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8.tflite) | COCO 2017 + PASCAL VOC 2012 | Int8 (tflite) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 806.12 | 8.73| 91.27 | 0 | v5.1.0 | OpenVX |
94
- | [DeepLabV3 MobileNetv2 ASPPv1 mixed precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8_f32.tflite) | COCO 2017 + PASCAL VOC 2012 | Int8 & float32 (tflite) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 894.56 | 7.67 | 92.33 | 0 | v5.1.0 | OpenVX |
95
- | [DeepLabV3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_qdq_int8.onnx) | COCO 2017 + PASCAL VOC 2012 | Int8 (onnx) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 729.62 | 3.0 | 97.0 | 0 | v5.1.0| OpenVX |
96
 
97
  - **DeepLabV3 per tensor**:
98
  This model, which does not include ASPP (Atrous Spatial Pyramid Pooling), was downloaded from the TensorFlow DeepLabV3 page on [Kaggle](https://www.kaggle.com/models/tensorflow/deeplabv3/).
99
 
100
  - **The onnx DeepLabv3 MobileNetv2 ASPPv1 per channel**:
101
- The quantized TFLite model is derived from the DeepLabV3 float precision model. The ONNX quantized model is obtained by quantizing the DeepLabV3 float model using the [deeplab_v3_mobilenetv2_05_16_512_asppv1_onnx_config](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_onnx_config.yaml) YAML file.
102
 
103
  **Note:** These results were obtained using the exact YAML files mentioned above and a specific quantization set containing 4 images from the PASCAL VOC dataset with the following IDs:
104
  - 2008_004804
@@ -123,7 +123,7 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
123
  - 1449 validation images and masks
124
 
125
 
126
- **Please follow the [PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/datasets) to have more training masks (about 10,582) and a `trainaug.txt` file containing the IDs of the new training masks.**
127
 
128
 
129
  **COCO Dataset Details:**
@@ -134,29 +134,29 @@ Measures are done with default STM32Cube.AI configuration with enabled input / o
134
  Please note, that the following accuracies are obtained after training the model with the augmented Pascal VOC + COCO data and evaluated on Pascal VOC 2012 validation set (val.txt), and with a preprocessing resize with interpolation method 'bilinear'.
135
  Moreover, IoU are averaged on all classes including background.
136
 
137
- **Please use the [COCO 2017 PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/datasets/coco_2017_pascal_voc_2012) to create COCO 2017 + PASCAL VOC 2012 dataset to do the needed filtering. Only images containing one or more classes from the 21 Pascal VOC dataset classes should be used. Additionally, the masks need to be converted to the Pascal VOC masks format.**
138
 
139
  | Model Description | Resolution | Format | Accuracy | Averaed IoU |
140
  |--------------------------------------------------------------------------------------------------------------------------------------------------------------|------------|------------|----------|--------------|
141
  | [DeepLabV3 per tensor (no ASPP)](https://www.st.com/en/embedded-software/x-linux-ai.html) | 257x257x3 | Int8 (tflite)| 88.6% | 59.33% |
142
- | [Deeplabv3 MobileNetv2 ASPPv1 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1.h5) | 512x512x3 | Float | 93.29% | 73.44% |
143
- | [DeepLabv3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8.tflite) | 512x512x3 | Int8 (tflite) | 91.3% | 67.32% |
144
- | [DDeepLabv3 MobileNetv2 ASPPv1 mixed precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8_f32.tflite) | 512x512x3 | Int8/Float (tflite)| 92.83% | 71.93% |
145
- | [DeepLabv3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_qdq_int8.onnx) | 512x512x3 | Int8 (onnx) | 93.15%| 72.39% |
146
 
147
 
148
  ### Accuracy with Person COCO 2017 + PASCAL VOC 2012
149
 
150
- **Please use the [Person COCO 2017 PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/datasets/n_class_coco_2017_pascal_voc_2012) to create Pesron COCO 2017 + PASCAL VOC 2012 dataset.**
151
 
152
  | Models Description | Resolution | Format | Accuracy (%) | average IoU |
153
  |--------------------------------------------|-----------|---------------|--------------|-------------|
154
- | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2.h5) | 256x256x3 | TensorFlow | 94.65 % | 76.96 % |
155
- | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | 256x256x | ONNX | 94.57 % | 76.62 % |
156
- | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2.h5) | 320x320x3 | TensorFlow | 95.16 % | 79.04 % |
157
- | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) | 320x320x3 | ONNX | 94.98 % | 78.35 % |
158
- | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2.h5) | 416x416x3 | TensorFlow | 95.48 % | 80.62 % |
159
- | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | 416x416x3 | ONNX | 95.44 % | 80.36 % |
160
 
161
 
162
  Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)
 
70
 
71
  | Model | Dataset | Format | Resolution | Series | Internal RAM (KiB) | External RAM (KiB) | Weights Flash (KiB) | STM32Cube.AI version | STEdgeAI Core version |
72
  |------------|---------------|----------|------------|-----------|--------------|--------------|---------------|----------------------|-----------------------|
73
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 256x256x3 | STM32N6 | 2253.5 | 0.0 | 1001.25 | 10.0.0 | 2.0.0 |
74
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) |person COCO 2017 + PASCAL VOC 2012 | Int8 | 320x320x3 | STM32N6 | 2446.0 | 0.0 | 1000.41 | 10.0.0 | 2.0.0 |
75
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 416x416x3 | STM32N6 | 2743.5 | 2028.0 | 2721.19 | 10.0.0 | 2.0.0 |
76
 
77
 
78
 
 
81
 
82
  | Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STM32Cube.AI version | STEdgeAI Core version |
83
  |------------|---------------|----------|------------|------------------|------------------|---------------------|-------------|----------------------|-------------------------|
84
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 256x256x3 | STM32N6570-DK | NPU/MCU | 27.36 | 36.54 | 10.0.0 | 2.0.0 |
85
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 320x320x3 | STM32N6570-DK | NPU/MCU | 44.99 | 22.22 | 10.0.0 | 2.0.0 |
86
+ | [DeepLabv3 MobileNetv2 ASPPv2](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | person COCO 2017 + PASCAL VOC 2012 | Int8 | 416x416x3 | STM32N6570-DK | NPU/MCU | 191.91 | 5.21 | 10.0.0 | 2.0.0 |
87
 
88
 
89
  ### Reference **MPU** inference time based on COCO 2017 + PASCAL VOC 2012 segmentation dataset 21 classes and a derivative person dataset from it (see Accuracy for details on dataset)
90
  | Model | Dataset | Format | Resolution | Quantization | Board| Execution Engine | Frequency | Inference time (ms) | %NPU | %GPU | %CPU | X-LINUX-AI version |Framework |
91
  |----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------|--------|------------|----------------|-------------------|------------------|-----------|---------------------|-------|--------|------|--------------------|-----------------------|
92
  | [DeepLabV3 per tensor (no ASPP)](https://www.st.com/en/embedded-software/x-linux-ai.html) | COCO 2017 + PASCAL VOC 2012 | Int8 | 257x257x3 | per-tensor | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 52.75 | 99.2 | 0.80 | 0 | v5.1.0 | OpenVX | | | | | v5.1.0
93
+ | [DeepLabV3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8.tflite) | COCO 2017 + PASCAL VOC 2012 | Int8 (tflite) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 806.12 | 8.73| 91.27 | 0 | v5.1.0 | OpenVX |
94
+ | [DeepLabV3 MobileNetv2 ASPPv1 mixed precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8_f32.tflite) | COCO 2017 + PASCAL VOC 2012 | Int8 & float32 (tflite) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 894.56 | 7.67 | 92.33 | 0 | v5.1.0 | OpenVX |
95
+ | [DeepLabV3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_qdq_int8.onnx) | COCO 2017 + PASCAL VOC 2012 | Int8 (onnx) | 512x512x3 | per-channel ** | STM32MP257F-DK2 | NPU/GPU | 1500 MHz | 729.62 | 3.0 | 97.0 | 0 | v5.1.0| OpenVX |
96
 
97
  - **DeepLabV3 per tensor**:
98
  This model, which does not include ASPP (Atrous Spatial Pyramid Pooling), was downloaded from the TensorFlow DeepLabV3 page on [Kaggle](https://www.kaggle.com/models/tensorflow/deeplabv3/).
99
 
100
  - **The onnx DeepLabv3 MobileNetv2 ASPPv1 per channel**:
101
+ The quantized TFLite model is derived from the DeepLabV3 float precision model. The ONNX quantized model is obtained by quantizing the DeepLabV3 float model using the [deeplab_v3_mobilenetv2_05_16_512_asppv1_onnx_config](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_onnx_config.yaml) YAML file.
102
 
103
  **Note:** These results were obtained using the exact YAML files mentioned above and a specific quantization set containing 4 images from the PASCAL VOC dataset with the following IDs:
104
  - 2008_004804
 
123
  - 1449 validation images and masks
124
 
125
 
126
+ **Please follow the [PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/tree/main/semantic_segmentation/datasets) to have more training masks (about 10,582) and a `trainaug.txt` file containing the IDs of the new training masks.**
127
 
128
 
129
  **COCO Dataset Details:**
 
134
  Please note, that the following accuracies are obtained after training the model with the augmented Pascal VOC + COCO data and evaluated on Pascal VOC 2012 validation set (val.txt), and with a preprocessing resize with interpolation method 'bilinear'.
135
  Moreover, IoU are averaged on all classes including background.
136
 
137
+ **Please use the [COCO 2017 PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/tree/main/semantic_segmentation/datasets/coco_2017_pascal_voc_2012) to create COCO 2017 + PASCAL VOC 2012 dataset to do the needed filtering. Only images containing one or more classes from the 21 Pascal VOC dataset classes should be used. Additionally, the masks need to be converted to the Pascal VOC masks format.**
138
 
139
  | Model Description | Resolution | Format | Accuracy | Averaed IoU |
140
  |--------------------------------------------------------------------------------------------------------------------------------------------------------------|------------|------------|----------|--------------|
141
  | [DeepLabV3 per tensor (no ASPP)](https://www.st.com/en/embedded-software/x-linux-ai.html) | 257x257x3 | Int8 (tflite)| 88.6% | 59.33% |
142
+ | [Deeplabv3 MobileNetv2 ASPPv1 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1.h5) | 512x512x3 | Float | 93.29% | 73.44% |
143
+ | [DeepLabv3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8.tflite) | 512x512x3 | Int8 (tflite) | 91.3% | 67.32% |
144
+ | [DDeepLabv3 MobileNetv2 ASPPv1 mixed precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_int8_f32.tflite) | 512x512x3 | Int8/Float (tflite)| 92.83% | 71.93% |
145
+ | [DeepLabv3 MobileNetv2 ASPPv1 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_512/deeplab_v3_mobilenetv2_05_16_512_asppv1_qdq_int8.onnx) | 512x512x3 | Int8 (onnx) | 93.15%| 72.39% |
146
 
147
 
148
  ### Accuracy with Person COCO 2017 + PASCAL VOC 2012
149
 
150
+ **Please use the [Person COCO 2017 PASCAL VOC 2012 tutorial](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/tree/main/semantic_segmentation/datasets/n_class_coco_2017_pascal_voc_2012) to create Pesron COCO 2017 + PASCAL VOC 2012 dataset.**
151
 
152
  | Models Description | Resolution | Format | Accuracy (%) | average IoU |
153
  |--------------------------------------------|-----------|---------------|--------------|-------------|
154
+ | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2.h5) | 256x256x3 | TensorFlow | 94.65 % | 76.96 % |
155
+ | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_256/deeplab_v3_mobilenetv2_05_16_256_asppv2_qdq_int8.onnx) | 256x256x | ONNX | 94.57 % | 76.62 % |
156
+ | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2.h5) | 320x320x3 | TensorFlow | 95.16 % | 79.04 % |
157
+ | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_320/deeplab_v3_mobilenetv2_05_16_320_asppv2_qdq_int8.onnx) | 320x320x3 | ONNX | 94.98 % | 78.35 % |
158
+ | [Deeplabv3 MobileNetv2 ASPPv2 float precision](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2.h5) | 416x416x3 | TensorFlow | 95.48 % | 80.62 % |
159
+ | [DeepLabv3 MobileNetv2 ASPPv2 per channel](https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/semantic_segmentation/deeplab_v3ST_pretrainedmodel_public_dataset/person_coco_2017_pascal_voc_2012/deeplab_v3_mobilenetv2_05_16_416/deeplab_v3_mobilenetv2_05_16_416_asppv2_qdq_int8.onnx) | 416x416x3 | ONNX | 95.44 % | 80.36 % |
160
 
161
 
162
  Please refer to the stm32ai-modelzoo-services GitHub [here](https://github.com/STMicroelectronics/stm32ai-modelzoo-services)