From 7d01ab05c0ea905fe9a1b6b8b6f1b418fb9edddf Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 12:12:56 +0300 Subject: [PATCH 01/17] added nyu fcrn depth prediction model --- models/public/index.md | 1 + models/public/nyu-fcrn-tf/accuracy-check.yml | 20 ++++ models/public/nyu-fcrn-tf/model.yml | 38 +++++++ models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 106 ++++++++++++++++++ .../accuracy_checker/configs/nyu-fcrn-tf.yml | 1 + .../accuracy_checker/dataset_definitions.yml | 9 ++ 6 files changed, 175 insertions(+) create mode 100644 models/public/nyu-fcrn-tf/accuracy-check.yml create mode 100644 models/public/nyu-fcrn-tf/model.yml create mode 100644 models/public/nyu-fcrn-tf/nyu-fcrn-tf.md create mode 120000 tools/accuracy_checker/configs/nyu-fcrn-tf.yml diff --git a/models/public/index.md b/models/public/index.md index 8075003df9d..4e510d3a1e1 100644 --- a/models/public/index.md +++ b/models/public/index.md @@ -170,6 +170,7 @@ Since this task contains - in the general setting - some ambiguity, the resultin | Model Name | Implementation | OMZ Model Name | Accuracy | GFlops | mParams | | --------------------------- | -------------- | ----------------------------- | -------- | --------- | ------- | | midasnet | PyTorch\* | [midasnet](./midasnet/midasnet.md)| 7.5878| 207.4915 | 104.0814 | +| NYU_FCRN | TensorFlow\* | [nyu-fcrn-tf](./nyu-fcrn-tf/nyu-fcrn-tf.md)| 0.573 | 63.5421 | 34.5255 | ## Image Inpainting diff --git a/models/public/nyu-fcrn-tf/accuracy-check.yml b/models/public/nyu-fcrn-tf/accuracy-check.yml new file mode 100644 index 00000000000..e89163a459f --- /dev/null +++ b/models/public/nyu-fcrn-tf/accuracy-check.yml @@ -0,0 +1,20 @@ +models: + - name: nyu-fcrn-tf + launchers: + - framework: dlsdk + adapter: mono_depth + datasets: + - name: NYU_Depth_V2 + preprocessing: + - type: resize + use_pillow: true + dst_height: 228 + dst_width: 304 + postprocessing: + - type: resize_prediction_depth_map + metrics: + - type: rmse + - type: log10_error + name: log10 + - type: mape + name: rel diff --git a/models/public/nyu-fcrn-tf/model.yml b/models/public/nyu-fcrn-tf/model.yml new file mode 100644 index 00000000000..4e71f2c5830 --- /dev/null +++ b/models/public/nyu-fcrn-tf/model.yml @@ -0,0 +1,38 @@ +# Copyright (c) 2020 Intel Corporation +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +description: >- + NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset; + as described in the paper "Deeper Depth Prediction with Fully Convolutional Residual Networks" + . + The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. + The model output is an inverse depth map that is defined up to an unknown scale factor. +task_type: monocular_depth_estimation +framework: tf +files: + - name: NYU_FCRN-checkpoint.zip + size: 472588519 + sha256: 9d97ed165c4a5b3f085eb83b8814de1e883c6348da60da4b2568ddd64bb2d5c4 + source: http://campar.in.tum.de/files/rupprecht/depthpred/NYU_FCRN-checkpoint.zip +postprocessing: + - $type: unpack_archive + format: zip + file: NYU_FCRN-checkpoint.zip +model_optimizer_args: + - --input=Placeholder + - --reverse_input_channels + - --input_shape=[1,228,304,3] + - --output=ConvPred/ConvPred + - --input_meta=$dl_dir/NYU_FCRN.ckpt.meta +license: https://raw.githubusercontent.com/iro-cp/FCRN-DepthPrediction/master/LICENSE diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md new file mode 100644 index 00000000000..8e00f817553 --- /dev/null +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -0,0 +1,106 @@ +# midasnet + +## Use Case and High-Level Description + +NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset; + as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373) + The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. + 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](https://github.com/iro-cp/FCRN-DepthPrediction). + + +## Specification + +| Metric | Value | +|-------------------|---------------| +| Type | Monodepth | +| GFLOPs | 63.5421 | +| MParams | 34.5255 | +| Source framework | TensorFlow\* | + +## Accuracy + +| Metric | Value | +| ------ | ----- | +| rmse | 0.573 | +| log10 | 0.055 | +| rel | 0.127 | + +## Input + +### Original Model + +Image, name - `Placeholder`, shape - `1,228,304,3`, format is `B,H,W,C` where: + +- `B` - batch size +- `C` - channel +- `H` - height +- `W` - width + +Channel order is `RGB`. + + +### Converted Model + +Image, name - `Placeholder`, shape - `1,3,228,304`, format is `B,C,H,W` where: + +- `B` - batch size +- `C` - channel +- `H` - height +- `W` - width + +Channel order is `BGR`. + +## Output + +### Original Model + +Inverse depth map, name - `ConvPred/ConvPred`, shape - `1,128,160`, format is `B,H,W` where: + +- `B` - batch size +- `H` - height +- `W` - width + +Inverse depth map is defined up to an unknown scale factor. + +### Converted Model + +Inverse depth map, name - `ConvPred/ConvPred`, shape - `1,128,160`, format is `B,H,W` where: + +- `B` - batch size +- `C` - channels +- `H` - height +- `W` - width + +Inverse depth map is defined up to an unknown scale factor. + +## Legal Information + +The original model is released under the following [license](https://raw.githubusercontent.com/iro-cp/FCRN-DepthPrediction/master/LICENSE): + +``` +Copyright (c) 2016, Iro Laina +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +* Redistributions of source code must retain the above copyright notice, this + list of conditions and the following disclaimer. + +* Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +``` + +[*] Other names and brands may be claimed as the property of others. diff --git a/tools/accuracy_checker/configs/nyu-fcrn-tf.yml b/tools/accuracy_checker/configs/nyu-fcrn-tf.yml new file mode 120000 index 00000000000..78e4ef2c93a --- /dev/null +++ b/tools/accuracy_checker/configs/nyu-fcrn-tf.yml @@ -0,0 +1 @@ +../../../models/public/nyu-fcrn-tf/accuracy-check.yml \ No newline at end of file diff --git a/tools/accuracy_checker/dataset_definitions.yml b/tools/accuracy_checker/dataset_definitions.yml index 644a96b7a33..3a576e469c7 100644 --- a/tools/accuracy_checker/dataset_definitions.yml +++ b/tools/accuracy_checker/dataset_definitions.yml @@ -957,3 +957,12 @@ datasets: - type: pad_with_eos eos_index: 1 sequence_len: 192 + + - name: NYU_Depth_V2 + data_source: nyudepthv2/val/converted/images + additional_data_source: nyudepthv2/val/converted/depth + annotation_conversion: + converter: nyu_depth_v2 + images_dir: nyudepthv2/val/converted/images + depth_map_dir: nyudepthv2/val/converted/depth + data_dir: nyudepthv2/val/official From 46997933130e175032ba391eb13816d0f55eb74b Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 12:25:20 +0300 Subject: [PATCH 02/17] extend metrics info --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index 8e00f817553..ec373625f8f 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -21,9 +21,12 @@ NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 datas | Metric | Value | | ------ | ----- | -| rmse | 0.573 | +| [rmse](https://en.wikipedia.org/wiki/Root-mean-square_deviation) | 0.573 | | log10 | 0.055 | -| rel | 0.127 | +| [rel](https://en.wikipedia.org/wiki/Mean_absolute_percentage_error) | 0.127 | + +Accuracy numbers obtained on NUY Depth V2 dataset. log10 is logarithmic absolute error, defined as abs(log10(gt) - log10(pred)), where gt - ground truth depth map, pred - predicted depth map. + ## Input From c3fed212eb65163898657b09f16375ad6cd3a91a Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:23:04 +0300 Subject: [PATCH 03/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index ec373625f8f..f009adcc69a 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -1,4 +1,4 @@ -# midasnet +# nyu-fcrn-tf ## Use Case and High-Level Description From 386f59e9e77c43ada407ef8fb6ef6f87a507964c Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:23:18 +0300 Subject: [PATCH 04/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index f009adcc69a..a9798a5c6d7 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -3,7 +3,7 @@ ## Use Case and High-Level Description NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset; - as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373) + as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373). The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. 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](https://github.com/iro-cp/FCRN-DepthPrediction). From c8ae888aad68aee9ce170cf5ce759d6ac4ec6e5e Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:24:05 +0300 Subject: [PATCH 05/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 1 - 1 file changed, 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index a9798a5c6d7..5eba10f24ff 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -41,7 +41,6 @@ Image, name - `Placeholder`, shape - `1,228,304,3`, format is `B,H,W,C` where: Channel order is `RGB`. - ### Converted Model Image, name - `Placeholder`, shape - `1,3,228,304`, format is `B,C,H,W` where: From 900bb94ddeaf945fad00e042a2ee813785317159 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:24:19 +0300 Subject: [PATCH 06/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 1 - 1 file changed, 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index 5eba10f24ff..4c797b5b2f3 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -69,7 +69,6 @@ Inverse 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 size -- `C` - channels - `H` - height - `W` - width From d6fb1e0713c6ea973e95caf0e37fa9c43a5f2153 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:25:09 +0300 Subject: [PATCH 07/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index 4c797b5b2f3..403815b6ba7 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -2,7 +2,7 @@ ## Use Case and High-Level Description -NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset; +NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset, as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373). The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. 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](https://github.com/iro-cp/FCRN-DepthPrediction). From 1a052a6f533f5bec3018feedeadf2d50366f9735 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 13:25:17 +0300 Subject: [PATCH 08/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index 403815b6ba7..bafb85161c0 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -21,7 +21,7 @@ NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 datas | Metric | Value | | ------ | ----- | -| [rmse](https://en.wikipedia.org/wiki/Root-mean-square_deviation) | 0.573 | +| [RMSE](https://en.wikipedia.org/wiki/Root-mean-square_deviation) | 0.573 | | log10 | 0.055 | | [rel](https://en.wikipedia.org/wiki/Mean_absolute_percentage_error) | 0.127 | From e97c4729f2a429c6e06ba36bddfce9b4e50995cd Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 15:07:30 +0300 Subject: [PATCH 09/17] rewrite rel definition --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index bafb85161c0..da4d6881fff 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -23,9 +23,13 @@ NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 datas | ------ | ----- | | [RMSE](https://en.wikipedia.org/wiki/Root-mean-square_deviation) | 0.573 | | log10 | 0.055 | -| [rel](https://en.wikipedia.org/wiki/Mean_absolute_percentage_error) | 0.127 | +| rel | 0.127 | -Accuracy numbers obtained on NUY Depth V2 dataset. log10 is logarithmic absolute error, defined as abs(log10(gt) - log10(pred)), where gt - ground truth depth map, pred - predicted depth map. +Accuracy numbers obtained on NUY Depth V2 dataset. +`log10` is logarithmic absolute error, defined as `abs(log10(gt) - log10(pred))`, +where `gt` - ground truth depth map, `pred` - predicted depth map. +`rel` 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). ## Input From 12f00293f9ef0175f8c2599bebde2d84adfe3597 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 17:19:17 +0300 Subject: [PATCH 10/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index da4d6881fff..eacd30ad85b 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -26,7 +26,7 @@ NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 datas | rel | 0.127 | Accuracy numbers obtained on NUY Depth V2 dataset. -`log10` is logarithmic absolute error, defined as `abs(log10(gt) - log10(pred))`, +The `log10` metric is logarithmic absolute error, defined as `abs(log10(gt) - log10(pred))`, where `gt` - ground truth depth map, `pred` - predicted depth map. `rel` 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). From 4d091bf4e36f826a6cf9fdb43284cb6fdd0901d2 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 17:19:30 +0300 Subject: [PATCH 11/17] Update models/public/nyu-fcrn-tf/nyu-fcrn-tf.md Co-authored-by: Eduard Zamaliev --- models/public/nyu-fcrn-tf/nyu-fcrn-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md index eacd30ad85b..a13c06bd9c4 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md @@ -28,7 +28,7 @@ NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 datas 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. -`rel` is relative absolute error defined as absolute error normalized on ground truth depth map values +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). From 74baec660019721665a51568c88e5777013f8a4b Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Mon, 19 Oct 2020 17:29:33 +0300 Subject: [PATCH 12/17] update demo --- demos/python_demos/monodepth_demo/models.lst | 1 + demos/python_demos/monodepth_demo/monodepth_demo.py | 2 +- 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/demos/python_demos/monodepth_demo/models.lst b/demos/python_demos/monodepth_demo/models.lst index b41b3069967..54c594746ea 100644 --- a/demos/python_demos/monodepth_demo/models.lst +++ b/demos/python_demos/monodepth_demo/models.lst @@ -1,2 +1,3 @@ # This file can be used with the --list option of the model downloader. midasnet +nyu-rfcn-tf diff --git a/demos/python_demos/monodepth_demo/monodepth_demo.py b/demos/python_demos/monodepth_demo/monodepth_demo.py index 483f32ff5eb..ff226bc8787 100755 --- a/demos/python_demos/monodepth_demo/monodepth_demo.py +++ b/demos/python_demos/monodepth_demo/monodepth_demo.py @@ -73,7 +73,7 @@ def main(): # processing output blob log.info("processing output blob") - disp = res[out_blob][0] + disp = np.squeeze(res[out_blob][0]) # resize disp to input resolution disp = cv2.resize(disp, (input_width, input_height), cv2.INTER_CUBIC) From ddc7a62e487fe0fc1adaa6d149ad349c74364707 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Tue, 20 Oct 2020 08:50:56 +0300 Subject: [PATCH 13/17] rename model --- demos/python_demos/monodepth_demo/models.lst | 2 +- .../accuracy-check.yml | 2 +- .../fcrn-dp-nyu-depth-v2-tf.md} | 4 ++-- .../public/{nyu-fcrn-tf => fcrn-dp-nyu-depth-v2-tf}/model.yml | 4 ++-- models/public/index.md | 2 +- tools/accuracy_checker/configs/fcrn-dp-nyu-depth-v2-tf.yml | 1 + tools/accuracy_checker/configs/nyu-fcrn-tf.yml | 1 - 7 files changed, 8 insertions(+), 8 deletions(-) rename models/public/{nyu-fcrn-tf => fcrn-dp-nyu-depth-v2-tf}/accuracy-check.yml (93%) rename models/public/{nyu-fcrn-tf/nyu-fcrn-tf.md => fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md} (96%) rename models/public/{nyu-fcrn-tf => fcrn-dp-nyu-depth-v2-tf}/model.yml (91%) create mode 120000 tools/accuracy_checker/configs/fcrn-dp-nyu-depth-v2-tf.yml delete mode 120000 tools/accuracy_checker/configs/nyu-fcrn-tf.yml diff --git a/demos/python_demos/monodepth_demo/models.lst b/demos/python_demos/monodepth_demo/models.lst index 2a7aa5d4bae..2d74ba1dd0d 100644 --- a/demos/python_demos/monodepth_demo/models.lst +++ b/demos/python_demos/monodepth_demo/models.lst @@ -1,3 +1,3 @@ # This file can be used with the --list option of the model downloader. midasnet -nyu-fcrn-tf +fcrn-dp-nyu-depth-v2-tf diff --git a/models/public/nyu-fcrn-tf/accuracy-check.yml b/models/public/fcrn-dp-nyu-depth-v2-tf/accuracy-check.yml similarity index 93% rename from models/public/nyu-fcrn-tf/accuracy-check.yml rename to models/public/fcrn-dp-nyu-depth-v2-tf/accuracy-check.yml index e89163a459f..55846148e57 100644 --- a/models/public/nyu-fcrn-tf/accuracy-check.yml +++ b/models/public/fcrn-dp-nyu-depth-v2-tf/accuracy-check.yml @@ -1,5 +1,5 @@ models: - - name: nyu-fcrn-tf + - name: fcrn-dp-nyu-depth-v2-tf launchers: - framework: dlsdk adapter: mono_depth diff --git a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md similarity index 96% rename from models/public/nyu-fcrn-tf/nyu-fcrn-tf.md rename to models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md index a13c06bd9c4..7afe2c5564f 100644 --- a/models/public/nyu-fcrn-tf/nyu-fcrn-tf.md +++ b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md @@ -1,8 +1,8 @@ -# nyu-fcrn-tf +# fcrn-dp-nyu-depth-v2-tf ## Use Case and High-Level Description -NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset, +FCRN ResNet50 UpProj is a model for monocular depth estimation trained by NYU Depth V2 dataset, as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373). The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. 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](https://github.com/iro-cp/FCRN-DepthPrediction). diff --git a/models/public/nyu-fcrn-tf/model.yml b/models/public/fcrn-dp-nyu-depth-v2-tf/model.yml similarity index 91% rename from models/public/nyu-fcrn-tf/model.yml rename to models/public/fcrn-dp-nyu-depth-v2-tf/model.yml index 4e71f2c5830..6a4727b67bb 100644 --- a/models/public/nyu-fcrn-tf/model.yml +++ b/models/public/fcrn-dp-nyu-depth-v2-tf/model.yml @@ -13,9 +13,9 @@ # limitations under the License. description: >- - NYU-FCRN is a model for monocular depth estimation trained by NYU Depth V2 dataset; + FCRN ResNet50 UpProj is a model for monocular depth estimation trained by NYU Depth V2 dataset; as described in the paper "Deeper Depth Prediction with Fully Convolutional Residual Networks" - . + . The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. The model output is an inverse depth map that is defined up to an unknown scale factor. task_type: monocular_depth_estimation diff --git a/models/public/index.md b/models/public/index.md index 4e510d3a1e1..3dd456a11d5 100644 --- a/models/public/index.md +++ b/models/public/index.md @@ -170,7 +170,7 @@ Since this task contains - in the general setting - some ambiguity, the resultin | Model Name | Implementation | OMZ Model Name | Accuracy | GFlops | mParams | | --------------------------- | -------------- | ----------------------------- | -------- | --------- | ------- | | midasnet | PyTorch\* | [midasnet](./midasnet/midasnet.md)| 7.5878| 207.4915 | 104.0814 | -| NYU_FCRN | TensorFlow\* | [nyu-fcrn-tf](./nyu-fcrn-tf/nyu-fcrn-tf.md)| 0.573 | 63.5421 | 34.5255 | +| FCRN ResNet50 Upproj | TensorFlow\* | [fcrn-dp-nyu-depth-v2-tf](./fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md)| 0.573 | 63.5421 | 34.5255 | ## Image Inpainting diff --git a/tools/accuracy_checker/configs/fcrn-dp-nyu-depth-v2-tf.yml b/tools/accuracy_checker/configs/fcrn-dp-nyu-depth-v2-tf.yml new file mode 120000 index 00000000000..6879bca328e --- /dev/null +++ b/tools/accuracy_checker/configs/fcrn-dp-nyu-depth-v2-tf.yml @@ -0,0 +1 @@ +../../../models/public/fcrn-dp-nyu-depth-v2-tf/accuracy-check.yml \ No newline at end of file diff --git a/tools/accuracy_checker/configs/nyu-fcrn-tf.yml b/tools/accuracy_checker/configs/nyu-fcrn-tf.yml deleted file mode 120000 index 78e4ef2c93a..00000000000 --- a/tools/accuracy_checker/configs/nyu-fcrn-tf.yml +++ /dev/null @@ -1 +0,0 @@ -../../../models/public/nyu-fcrn-tf/accuracy-check.yml \ No newline at end of file From 326efa5a805a760f0d07b1e7bfedec09a58a47af Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Tue, 20 Oct 2020 18:51:43 +0300 Subject: [PATCH 14/17] Update models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md Co-authored-by: Roman Donchenko --- .../public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md index 7afe2c5564f..01211ffa575 100644 --- a/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md +++ b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md @@ -2,8 +2,8 @@ ## Use Case and High-Level Description -FCRN ResNet50 UpProj is a model for monocular depth estimation trained by NYU Depth V2 dataset, - as described in the paper [Deeper Depth Prediction with Fully Convolutional Residual Networks](https://arxiv.org/abs/1606.00373). +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](https://arxiv.org/abs/1606.00373), where it is referred to as ResNet-UpProj. The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. 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](https://github.com/iro-cp/FCRN-DepthPrediction). From d86ab173b4745c105b5364a13f4c5c232ead9918 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Tue, 20 Oct 2020 18:52:11 +0300 Subject: [PATCH 15/17] Update models/public/index.md Co-authored-by: Roman Donchenko --- models/public/index.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/index.md b/models/public/index.md index 3dd456a11d5..a5dbbe144b9 100644 --- a/models/public/index.md +++ b/models/public/index.md @@ -170,7 +170,7 @@ Since this task contains - in the general setting - some ambiguity, the resultin | Model Name | Implementation | OMZ Model Name | Accuracy | GFlops | mParams | | --------------------------- | -------------- | ----------------------------- | -------- | --------- | ------- | | midasnet | PyTorch\* | [midasnet](./midasnet/midasnet.md)| 7.5878| 207.4915 | 104.0814 | -| FCRN ResNet50 Upproj | TensorFlow\* | [fcrn-dp-nyu-depth-v2-tf](./fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md)| 0.573 | 63.5421 | 34.5255 | +| FCRN ResNet50-Upproj | TensorFlow\* | [fcrn-dp-nyu-depth-v2-tf](./fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md)| 0.573 | 63.5421 | 34.5255 | ## Image Inpainting From 4615b9fce240f230f251ae5a2241a7076dc8e9e1 Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Tue, 20 Oct 2020 18:52:42 +0300 Subject: [PATCH 16/17] Update models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md Co-authored-by: Roman Donchenko --- .../public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md index 01211ffa575..2dd494f53bd 100644 --- a/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md +++ b/models/public/fcrn-dp-nyu-depth-v2-tf/fcrn-dp-nyu-depth-v2-tf.md @@ -4,7 +4,7 @@ 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](https://arxiv.org/abs/1606.00373), where it is referred to as ResNet-UpProj. - The model input is a blob that consists of a single image of "1x228x304x3" in RGB order. + 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](https://github.com/iro-cp/FCRN-DepthPrediction). From ab0712f3d12d1b7e9aab4311dd87f9c251e42b5d Mon Sep 17 00:00:00 2001 From: Ekaterina Aidova Date: Tue, 20 Oct 2020 18:53:52 +0300 Subject: [PATCH 17/17] Update models.lst --- demos/python_demos/monodepth_demo/models.lst | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/demos/python_demos/monodepth_demo/models.lst b/demos/python_demos/monodepth_demo/models.lst index 2d74ba1dd0d..12e2402681c 100644 --- a/demos/python_demos/monodepth_demo/models.lst +++ b/demos/python_demos/monodepth_demo/models.lst @@ -1,3 +1,3 @@ # This file can be used with the --list option of the model downloader. -midasnet fcrn-dp-nyu-depth-v2-tf +midasnet