geo-deep-learning
geo-deep-learning copied to clipboard
NRCan
Deep learning applied to georeferenced datasets
inference.py: keep raw heatmap to set confidence level for extracted features after post-processing
Inference values for each before using argmax() function will be useful for technicians for sort extracted features by confidence levels. Our inference script should let use decide to output the...
As mentionned by @Dan-Eli about a year ago, it would clean out the code a bit to refactor [visualization.py](https://github.com/NRCan/geo-deep-learning/blob/6852729d0ec03c0c2a6465b7a68039181c8bcbda/utils/visualization.py) as a Class (e.g. Visualizer). To make this refactorization worth while,...
Even though the ultimate goal of the extraction process is to have vector features, we should write our raster predictions in COGs. For one thing, data cube environments can be...
[Torch.cuda...](https://pytorch.org/docs/stable/cuda.html) utilities has everything we need to replace the pynvml utilities and dependency. 2 examples: - device count: [pynvml](https://github.com/NRCan/geo-deep-learning/blob/caf1223a8918cb5e9dc44e783e0ecc90c5aa9f0d/utils/utils.py#L115) vs [torch.cuda](https://pytorch.org/docs/stable/generated/torch.cuda.device_count.html) - get free memory: [pynvml](https://github.com/NRCan/geo-deep-learning/blob/caf1223a8918cb5e9dc44e783e0ecc90c5aa9f0d/utils/utils.py#L160) vs [torch.cuda](https://github.com/pytorch/pytorch/blob/b505adbb09e45d667519954440aaec0e6f216a1c/torch/cuda/memory.py#L569)
It can be confusing to understand where and how padding is done in inference. For example, [pad](https://github.com/NRCan/geo-deep-learning/blob/develop/inference_segmentation.py#L171) and [dist_samples](https://github.com/NRCan/geo-deep-learning/blob/develop/inference_segmentation.py#L173) are duplicate variables. Also, the padding value is currently hardcoded. It...
https://pytorch.org/docs/stable/amp.html Tests with current inference script show that by simply wrapping the `outputs = model(inputs)` call in a context manager `with autocast(device_type=device.type)` speeds up inference by about 15% (tested on...
Describe DL model characteristics -- e.g. metadata like what data we used for training, feature types that can be extracted -- using the emerging [STAC ml-model extension](https://github.com/stac-extensions/ml-model). - [ ]...
The inference module needs prioritization to create unit tests. Among the tests, one will have to check the "reading from STAC item" feature from #222
As models become more class-specific, the possibility to create a single inference from multiple models (ex.: combination of 4 class-specific models or 1 class-specific model and 1 multi-class model) becomes...
For examples, see [geo_sim_processing's github workflow](https://github.com/NRCan/geo_sim_processing/blob/master/.github/workflows/test_plugin.yaml) Also, see [Github Action example](https://github.com/NRCan/geo-deep-learning/new/develop?filename=.github%2Fworkflows%2Fdocker-image.yml&workflow_template=docker-image). Related to #143 and #228