Try out our models in Google Colab on your own images!
This repo includes the semantic segmentation pre-trained models, training and inference code for the paper:
MSeg: A Composite Dataset for Multi-domain Semantic Segmentation (CVPR 2020, Official Repo) [PDF]
John Lambert*,
Zhuang Liu*,
Ozan Sener,
James Hays,
Vladlen Koltun
Presented at CVPR 2020. Link to MSeg Video (3min)
This repo is the second of 4 repos that introduce our work. It provides utilities to train semantic segmentation models, using a HRNet-W48 or PSPNet backbone, sufficient to train a winning entry on the WildDash benchmark.
mseg-api
: utilities to download the MSeg dataset, prepare the data on disk in a unified taxonomy, on-the-fly mapping to a unified taxonomy during training.
Two additional repos will be introduced in July 2020:
mseg-panoptic
: provides Panoptic-FPN and Mask-RCNN training, based on Detectron2mseg-mturk
: provides utilities to perform large-scale Mechanical Turk re-labeling
First, install the mseg
module from mseg-api
Second, install the apex
module. NVIDIA's apex
is a library with Pytorch extensions. If your Pytorch version doesn't match the apex
version, this is likely fine, and you can comment out the bare metal version check in their setup.py
. Make sure your CUDA_HOME
is also set by running echo $CUDA_HOME
, it should be something like /usr/local/cuda-10.2
.
git clone https://github.com/NVIDIA/apex
cd apex
pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
-
mseg_semantic
can be installed as a python package usingpip install -e /path_to_root_directory_of_the_repo/
Make sure that you can run python -c "import mseg_semantic; print('hello world')"
in python, and you are good to go!
Each model is 528 MB in size. We provide download links and testing results (single-scale inference) below:
Abbreviated Dataset Names: VOC = PASCAL VOC, PC = PASCAL Context, WD = WildDash, SN = ScanNet
Model | Training Set | Training Taxonomy |
VOC mIoU |
PC mIoU |
CamVid mIoU |
WD mIoU |
KITTI mIoU |
SN mIoU |
h. mean | Download Link |
---|---|---|---|---|---|---|---|---|---|---|
MSeg (1M) | MSeg train | Universal | 70.7 | 42.7 | 83.3 | 62.0 | 67.0 | 48.2 | 59.2 | Google Drive |
MSeg (3M)-480p | MSeg train |
Universal | 76.4 | 45.9 | 81.2 | 62.7 | 68.2 | 49.5 | 61.2 | Google Drive |
MSeg (3M)-720p | MSeg train |
Universal | 74.7 | 44.0 | 83.5 | 60.4 | 67.9 | 47.7 | 59.8 | Google Drive |
MSeg (3M)-1080p | MSeg train |
Universal | 72.0 | 44.0 | 84.5 | 59.9 | 66.5 | 49.5 | 59.8 | Google Drive |
We show how to perform inference here in our Google Colab.
Multi-scale inference greatly improves the smoothness of predictions, therefore our demos scripts use multi-scale config by default. While we train at 1080p, our predictions are often visually better when we feed in test images at 360p resolution.
If you have video input, and you would like to make predictions on each frame in the universal taxonomy, please set:
input_file=/path/to/my/video.mp4
If you have a set of images in a directory, and you would like to make a prediction in the universal taxonomy for each image, please set:
input_file=/path/to/my/directory
If you have as input a single image, and you would like to make a prediction in the universal taxonomy, please set:
input_file=/path/to/my/image
Now, run our demo script:
model_name=mseg-3m
model_path=/path/to/downloaded/model/from/google/drive
config=mseg_semantic/config/test/default_config_360_ms.yaml
python -u mseg_semantic/tool/universal_demo.py \
--config=${config} model_name ${model_name} model_path ${model_path} input_file ${input_file}
To compute mIoU scores on all train and test datasets, run the following (single-scale inference):
cd mseg_semantic/scripts
./eval_models.sh
This will launch several hundred SLURM jobs, each of the following form:
python mseg_semantic/tool/test_universal_tax.py --config=$config_path dataset $dataset_name model_path $model_path model_name $model_name
Please expect this to take many hours, depending upon your SLURM cluster capacity.
"Oracle" models are trained in a test dataset's taxonomy, on its train split. To compute mIoU scores on the test dataset's val or test set, please run the following:
This will launch 5 SLURM jobs, each of the following form
python mseg_semantic/tool/test_oracle_tax.py
If you find this code useful for your research, please cite:
@InProceedings{MSeg_2020_CVPR,
author = {Lambert, John and Liu, Zhuang and Sener, Ozan and Hays, James and Koltun, Vladlen},
title = {{MSeg}: A Composite Dataset for Multi-domain Semantic Segmentation},
booktitle = {Computer Vision and Pattern Recognition (CVPR)},
year = {2020}
}
Many thanks to Hengshuang Zhao for his semseg repo, which we've based much of this repository off of.
Below we report the performance of individually-trained models that serve as baselines. Inference is performed at single-scale below.
You can obtain the following table by running
python collect_results.py --regime zero_shot --scale ss --output_format markdown
python collect_results.py --regime oracle --scale ss --output_format markdown
after ./eval_models.sh
:
Abbreviated Dataset Names: VOC = PASCAL VOC, PC = PASCAL Context, WD = WildDash, SN = ScanNet
Model | Training Set | Training Tax- onomy |
VOC mIoU |
PC mIoU |
CamVid mIoU |
WD mIoU |
KITTI mIoU |
SN mIoU |
h. mean | Download Link |
---|---|---|---|---|---|---|---|---|---|---|
ADE20K (1M) | ADE20K train | Universal | 35.4 | 23.9 | 52.6 | 38.6 | 41.6 | 42.9 | 36.9 | Google Drive |
BDD (1M) | BDD train | Universal | 14.4 | 7.1 | 70.7 | 52.2 | 54.5 | 1.4 | 6.1 | Google Drive |
Cityscapes (1M ) | Cityscapes train | Universal | 13.3 | 6.8 | 76.1 | 30.1 | 57.6 | 1.7 | 6.8 | Google Drive |
COCO (1M) | COCO train | Universal | 73.4 | 43.3 | 58.7 | 38.2 | 47.6 | 33.4 | 45.8 | Google Drive |
IDD (1M) | IDD train | Universal | 14.6 | 6.5 | 72.1 | 41.2 | 51.0 | 1.6 | 6.5 | Google Drive |
Mapillary (1M) | Mapillary train |
Universal | 22.5 | 13.6 | 82.1 | 55.4 | 67.7 | 2.1 | 9.3 | Google Drive |
SUN RGB-D (1M) | SUN RGBD train |
Universal | 10.0 | 4.3 | 0.1 | 1.9 | 1.1 | 42.6 | 0.3 | Google Drive |
MSeg (1M) | MSeg train. |
Universal | 70.7 | 42.7 | 83.3 | 62.0 | 67.0 | 48.2 | 59.2 | Google Drive |
MSeg Mix w/o relabeling (1M) | MSeg train. | Universal | 70.2 | 42.7 | 82.0 | 62.7 | 65.5 | 43.2 | 57.6 | Google Drive |
MGDA Baseline (1M) | MSeg train. | Universal | 68.5 | 41.7 | 82.9 | 61.1 | 66.7 | 46.7 | 58.0 | Google Drive |
MSeg (3M)-480p | MSeg train |
Universal | 76.4 | 45.9 | 81.2 | 62.7 | 68.2 | 49.5 | 61.2 | Google Drive |
MSeg (3M)-720p | MSeg train |
Universal | 74.7 | 44.0 | 83.5 | 60.4 | 67.9 | 47.7 | 59.8 | Google Drive |
MSeg (3M)-1080p | MSeg train |
Universal | 72.0 | 44.0 | 84.5 | 59.9 | 66.5 | 49.5 | 59.8 | Google Drive |
Naive Mix Baseline (1M) | MSeg train. |
Naive | Google Drive | |||||||
Oracle (1M) | 77.8 | 45.8 | 78.8 | -** | 58.4 | 62.3 | - |
**WildDash has no training set, so an "oracle" model cannot be trained.
Oracle Model Download Links
- VOC 2012 (1M) Model Google Drive
- PASCAL-Context (1M) Model Google Drive
- Camvid (1M) Model Google Drive
- KITTI (1M) Model Google Drive
- ScanNet-20 (1M) Model Google Drive
Note that the output number of classes for 7 of the models listed above will be identical (194 classes). These are the models that represent a single training dataset's performance -- ADE20K (1M), BDD (1M), Cityscapes (1M ), COCO (1M), IDD (1M), Mapillary (1M), SUN RGB-D (1M). When we train a baseline model on a single dataset, we train it in the universal taxonomy (w/ 194 classes). If we did not do so, we would need to specify 7*6=42 mappings (which would be unbelievably tedious and also fairly redundant) since we measure each's performance according to zero-shot cross-dataset generalization -- 7 training datasets with their own taxonomy, and each would need its own mapping to each of the 6 test sets.
By training each baseline that is trained on a single training dataset within the universal taxonomy, we are able to specify just 7+6=13 mappings in this table (each training dataset's taxonomy->universal taxonomy, and then universal taxonomy to each test dataset).
You can obtain the following table by running
python collect_results.py --regime training_datasets --scale ss --output_format markdown
after ./eval_models.sh
:
Model Name | COCO | ADE20k | Mapill | IDD | BDD | Citysca | SUN-RGBD | h. mean |
---|---|---|---|---|---|---|---|---|
COCO-1m | 52.7 | 19.1 | 28.4 | 31.1 | 44.9 | 46.9 | 29.6 | 32.4 |
ADE20K-1m | 14.6 | 45.6 | 24.2 | 26.8 | 40.7 | 44.3 | 36.0 | 28.7 |
Mapillary-1m | 7.0 | 6.2 | 53.0 | 50.6 | 59.3 | 71.9 | 0.3 | 1.7 |
IDD-1m | 3.2 | 3.0 | 24.6 | 64.9 | 42.4 | 48.0 | 0.4 | 2.3 |
BDD-1m | 3.8 | 4.2 | 23.2 | 32.3 | 63.4 | 58.1 | 0.3 | 1.6 |
Cityscapes-1m | 3.4 | 3.1 | 22.1 | 30.1 | 44.1 | 77.5 | 0.2 | 1.2 |
SUN RGBD-1m | 3.4 | 7.0 | 1.1 | 1.0 | 2.2 | 2.6 | 43.0 | 2.1 |
MSeg-1m | 50.7 | 45.7 | 53.1 | 65.3 | 68.5 | 80.4 | 50.3 | 57.1 |
MSeg-1m-w/o relabeling | 50.4 | 45.4 | 53.1 | 65.1 | 66.5 | 79.5 | 49.9 | 56.6 |
MSeg-MGDA-1m | 48.1 | 43.7 | 51.6 | 64.1 | 67.2 | 78.2 | 49.9 | 55.4 |
MSeg-3m-480p | 56.1 | 49.6 | 53.5 | 64.5 | 67.8 | 79.9 | 49.2 | 58.5 |
MSeg-3m-720p | 53.3 | 48.2 | 53.5 | 64.8 | 68.6 | 79.8 | 49.3 | 57.8 |
MSeg-3m-1080p | 53.6 | 49.2 | 54.9 | 66.3 | 69.1 | 81.5 | 50.1 | 58.8 |
We use the HRNetV2-W48 architecture. All images are resized to 1080p (shorter size=1080) at training time before a crop is taken.
We run inference with the shorter side of each test image at three resolutions (360p, 720p, 1080p), and take the max among these 3 possible resolutions. Note that in the original semseg repo, the author specifies the longer side of an image, whereas we specify the shorter side. Batch size is set to 35.
We generally follow the recommendations of Zhao et al.: Our data augmentation consists of random scaling in the range [0.5,2.0], random rotation in the range [-10,10] degrees. We use SGD with momentum 0.9, weight decay of 1e-4. We use a polynomial learning rate with power 0.9. Base learning rate is set to 1e-2. An auxiliary cross-entropy (CE) loss is added to intermediate activations, a linear combination with weight 0.4. In our data, we use 255 as an ignore/unlabeled flag for the CE loss. We use Pytorch's Distributed Data Parallel (DDP) package for multiprocessing, with the NCCL backend. We use apex opt_level: 'O0' and use a crop size of 713x713, with synchronized BN.
Download the HRNet Backbone Model here from the original authors' OneDrive. We use 8 Quadro RTX 6000 cards, each w/ 24 GB of RAM, for training.
To run the unit tests, execute
pytest tests
All should pass. To run the integration tests, follow the instructions in the following 3 files, then run:
python test_test_oracle_tax.py
python test_test_universal_tax.py
python test_universal_demo.py
All should also pass.
Frequently Asked Questions (FAQ) (identical to FAQ on mseg-api
page)
Q: Do the weights include the model structure or it's just the weights? If the latter, which model do these weights refer to? Under the models
directory, there are several model implementations.
A: The pre-trained models follow the HRNet-W48 architecture. The model structure is defined in the code here. The saved weights provide a dictionary between keys (unique IDs for each weight identifying the corresponding layer/layer type) and values (the floating point weights).
Q: How is testing performed on the test datasets? In the paper you talk about "zero-shot transfer" -- how this is performed? Are the test dataset labels also mapped or included in the unified taxonomy? If you remapped the test dataset labels to the unified taxonomy, are the reported results the performances on the unified label space, or on each test dataset's original label space? How did you you obtain results on the WildDash dataset - which is evaluated by the server - when the MSeg taxonomy may be different from the WildDash dataset.
A: Regarding "zero-shot transfer", please refer to section "Using the MSeg taxonomy on a held-out dataset" on page 6 of our paper. This section describes how we hand-specify mappings from the unified taxonomy to each test dataset's taxonomy as a linear mapping (implemented here in mseg-api). All results are in the test dataset's original label space (i.e. if WildDash expects class indices in the range [0,18] per our names_list, our testing script uses the TaxonomyConverter
transform_predictions_test()
functionality to produce indices in that range, remapping probabilities.
Q: Why don't indices in MSeg_master.tsv
match the training indices in individual datasets? For example, for the road class: In idd-39, road has index 0, but in idd-39-relabeled, road has index 19. It is index 7 in cityscapes-34. The cityscapes-19-relabeled index road is 11. As far as I can tell, ultimately the 'MSeg_Master.tsv' file provides the final mapping to the MSeg label space. But here, the road class seems to have an index of 98, which is neither 19 nor 11.
A: Indeed, unified taxonomy class index 98 represents "road". But we use the TaxonomyConverter to accomplish the mapping on the fly from idd-39-relabeled to the unified/universal taxonomy (we use the terms "unified" and "universal" interchangeably). This is done by adding a transform in the training loop that calls TaxonomyConverter.transform_label()
on the fly. You can see how that transform is implemented here in mseg-semantic
.
Q: When testing, but there are test classes that are not in the unified taxonomy (e.g. Parking, railtrack, bridge etc. in WildDash), how do you produce predictions for that class? I understand you map the predictions with a binary matrix. But what do you do when there's no one-to-one correspondence?
A: WildDash v1 uses the 19-class taxonomy for evaluation, just like Cityscapes. So we use the following script to remap the 34-class taxonomy to 19-class taxonomy for WildDash for testing inference and submission. You can see how Cityscapes evaluates just 19 of the 34 classes here in the evaluation script and in the taxonomy definition. However, bridge and rail track are actually included in our unified taxonomy, as you’ll see in MSeg_master.tsv.
Q: How are datasets images read in for training/inference? Should I use the dataset_apis
from mseg-api
?
A: The dataset_apis
from mseg-api
are not for training or inference. They are purely for generating the MSeg dataset labels on disk. We read in the datasets using mseg_semantic/utils/dataset.py
and then remap them to the universal space on the fly.