This is a model for monocular depth estimation trained on the NYU Depth V2 dataset, as described in the paper Deeper Depth Prediction with Fully Convolutional Residual Networks, where it is referred to as ResNet-UpProj. The model input is a single color image. The model output is an inverse depth map that is defined up to an unknown scale factor. More details can be found in the following repository.
Metric | Value |
---|---|
Type | Monodepth |
GFLOPs | 63.5421 |
MParams | 34.5255 |
Source framework | TensorFlow* |
Metric | Value |
---|---|
RMSE | 0.573 |
log10 | 0.055 |
rel | 0.127 |
Accuracy numbers obtained on NUY Depth V2 dataset. The log10
metric is logarithmic absolute error, defined as abs(log10(gt) - log10(pred))
, where gt
- ground truth depth map, pred
- predicted depth map. The rel
metric is relative absolute error defined as absolute error normalized on ground truth depth map values (abs(gt - pred) / gt
, where gt
- ground truth depth map, pred
- predicted depth map).
Image, name - Placeholder
, shape - 1,228,304,3
, format is B,H,W,C
where:
B
- batch sizeC
- channelH
- heightW
- widthChannel order is RGB
.
Image, name - Placeholder
, shape - 1,3,228,304
, format is B,C,H,W
where:
B
- batch sizeC
- channelH
- heightW
- widthChannel order is BGR
.
Inverse depth map, name - ConvPred/ConvPred
, shape - 1,128,160
, format is B,H,W
where:
B
- batch sizeH
- heightW
- widthInverse depth map is defined up to an unknown scale factor.
Inverse depth map, name - ConvPred/ConvPred
, shape - 1,128,160
, format is B,H,W
where:
B
- batch sizeH
- heightW
- widthInverse depth map is defined up to an unknown scale factor.
You can download models and if necessary convert them into Inference Engine format using the Model Downloader and other automation tools as shown in the examples below.
An example of using the Model Downloader:
An example of using the Model Converter:
The original model is released under the following license:
[*] Other names and brands may be claimed as the property of others.