{ "id": "corsa_compression", "links": [ { "href": "https://openeo-dev.vito.be/openeo/1.2/processes/u:63d26122aec1fc5ae2c16a045570e4d90e89fb3a51081c0d9d6047c270f72940@egi.eu/corsa_compression", "rel": "canonical", "title": "Public URL for user-defined process 'corsa_compression'" } ], "parameters": [ { "description": "A bounding box, which may include a vertical axis (see `base` and `height`).", "name": "spatial_extent", "schema": { "properties": { "base": { "default": null, "description": "Base (optional, lower left corner, coordinate axis 3).", "type": [ "number", "null" ] }, "crs": { "anyOf": [ { "examples": [ 3857 ], "minimum": 1000, "subtype": "epsg-code", "title": "EPSG Code", "type": "integer" }, { "subtype": "wkt2-definition", "title": "WKT2", "type": "string" } ], "default": 4326, "description": "Coordinate reference system of the extent, specified as as [EPSG code](http://www.epsg-registry.org/) or [WKT2 CRS string](http://docs.opengeospatial.org/is/18-010r7/18-010r7.html). Defaults to `4326` (EPSG code 4326) unless the client explicitly requests a different coordinate reference system." }, "east": { "description": "East (upper right corner, coordinate axis 1).", "type": "number" }, "height": { "default": null, "description": "Height (optional, upper right corner, coordinate axis 3).", "type": [ "number", "null" ] }, "north": { "description": "North (upper right corner, coordinate axis 2).", "type": "number" }, "south": { "description": "South (lower left corner, coordinate axis 2).", "type": "number" }, "west": { "description": "West (lower left corner, coordinate axis 1).", "type": "number" } }, "required": [ "west", "south", "east", "north" ], "subtype": "bounding-box", "type": "object" } }, { "default": [ "2023-05-01", "2023-05-01" ], "description": "Date expressed as an interval, all dates for which data is available will be processed.", "name": "temporal_extent", "optional": true, "schema": { "subtype": "temporal-interval", "type": "array" } } ], "process_graph": { "applyneighborhood1": { "arguments": { "data": { "from_node": "loadcollection1" }, "overlap": [ { "dimension": "x", "unit": "px", "value": 0 }, { "dimension": "y", "unit": "px", "value": 0 } ], "process": { "process_graph": { "runudf1": { "arguments": { "context": { "bands": [ "B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B11", "B12" ], "model_dir": "/data/users/Public/luytsa/corsa-compression/pretrain_BEN_10-20mbands_bicubic_512-128_onnx" }, "data": { "from_parameter": "data" }, "runtime": "Python", "udf": "import functools\nimport requests, zipfile, io\nimport os\nimport numpy as np\nimport logging\nimport xarray\nimport joblib\nfrom openeo.udf import XarrayDataCube\nfrom openeo.udf.debug import inspect\nfrom openeo.metadata import CollectionMetadata\n\nfrom typing import Dict\nimport traceback\nimport sys\n\n@functools.lru_cache(maxsize=25)\ndef load_onnx_deps():\n url = \"https://artifactory.vgt.vito.be:443/auxdata-public/openeo/onnx_dependencies_1.16.3.zip\"\n r = requests.get(url)\n z = zipfile.ZipFile(io.BytesIO(r.content))\n z.extractall(\"./onnx_deps\")\n\nload_onnx_deps()\nsys.path.insert(0, \"./onnx_deps\")\nimport onnxruntime as ort\n\n@functools.lru_cache(maxsize=25)\ndef load_ort_session(model_dir, session_options=None):\n session = ort.InferenceSession(f'{model_dir}/encoder.onnx', session_options=session_options)\n ort_input_name = session.get_inputs()[0].name\n return session, ort_input_name\n\n\n@functools.lru_cache(maxsize=25)\ndef load_scalers(scaler_dir, ordered_bands):\n scalers = {}\n for band in tuple(ordered_bands):\n scalers[band] = joblib.load(\n open(\n os.path.join(scaler_dir, f\"{band}_scaler.pkl\"), \"rb\"\n )\n )\n inspect(message=\"ULOGGING: loaded scaler band\", data=[band])\n return scalers\n\n# # # seems to have no effect :'(\ndef apply_metadata(input_metadata:CollectionMetadata, context:dict) -> CollectionMetadata:\n xstep = input_metadata.get('x','step')\n ystep = input_metadata.get('y','step')\n inspect(message=\"ULOGGING: x,y step\", data=[xstep, ystep])\n bands_out = ['level_0', 'level_1']\n renamed_bands = bands_out + [\"\"] * 8\n return input_metadata.rename_labels(dimension=\"bands\", target=renamed_bands)\n\n\n\n # xstep = input_metadata.get('x','step')\n # ystep = input_metadata.get('y','step')\n # inspect(message=\"ULOGGING: x,y step\", data=[xstep, ystep])\n # # # inspect(message=\"ULOGGING: metadata\", data=[input_metadata])\n # # print(f\"Metadata: {input_metadata}\")\n\n\n # new_metadata = {\n # \"x\": {\"type\": \"spatial\", \"axis\": \"x\", \"step\": xstep},\n # \"y\": {\"type\": \"spatial\", \"axis\": \"y\", \"step\": ystep},\n # \"t\": {\"type\": \"temporal\"}\n # }\n # return CollectionMetadata(new_metadata)\n\ndef process_window_onnx(input_data, model_dir, ort_sess_opt=None):\n \"\"\" Apply the model. Model expects [1, n_bands, 120, 120] input (already preprocessed) and returns [codebook_ids_level_0, codebook_ids_level_1]\"\"\"\n if ort_sess_opt is None:\n ort_sess_opt = ort.SessionOptions()\n ort_sess_opt.intra_op_num_threads = 1\n \n ort_session, ort_input_name = load_ort_session(model_dir, session_options=ort_sess_opt)\n ## apply model\n ort_inputs = {ort_input_name: input_data}\n ort_l0_vec, ort_l1_vec, ort_l0_ids, ort_l1_ids = ort_session.run(None, ort_inputs) # run() requires [output_names] and inputs arguments\n\n ort_l0_ids = ort_l0_ids.reshape(1,1,60,60)\n ort_l0_ids = ort_l0_ids.repeat(2, axis=2).repeat(2, axis=3)\n\n ort_l1_ids = ort_l1_ids.reshape(1,1,30,30)\n ort_l1_ids = ort_l1_ids.repeat(4, axis=2).repeat(4, axis=3)\n\n return np.concatenate([ort_l0_ids, ort_l1_ids], axis=1).astype(np.uint16)\n\ndef preprocess_datacube(input_data, scaler_dir, ordered_bands):\n scalers = load_scalers(scaler_dir, tuple(ordered_bands))\n\n input_data = input_data.data.astype(np.float32) #.squeeze(0)\n im_width = input_data.shape[2]\n im_height = input_data.shape[1]\n n_bands = input_data.shape[0]\n assert n_bands == len(scalers), f'found {len(scalers)} scalers, but {n_bands} bands were given as input. Should be equal.'\n\n for k, band in enumerate(ordered_bands):\n input_data[k] = np.clip(\n scalers[band]\n .transform(input_data[k].ravel().reshape(-1, 1))\n .reshape(im_height, im_width)\n ,-100,18000)\n \n # inspect(message=\"ULOGGING: after scaling\", data=[input_data])\n return input_data[np.newaxis, ...].astype(np.float32)\n\n\ndef apply_datacube(cube: XarrayDataCube, context: Dict) -> XarrayDataCube:\n\n \"\"\" main function \"\"\"\n\n try:\n ## get path to model, falling back to default value\n if len(context) > 0:\n model_dir = context.get(\"model_dir\", '/data/users/Public/luytsa/corsa-compression/onnx_ben_allbands')\n bands = context.get(\"bands\", ('B01', 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B09', 'B11', 'B12'))\n inspect(message=\"ULOGGING: path_model from context\", data=[model_dir, os.path.exists(model_dir)])\n else: # for local debugging\n model_dir = '/home/vito/luytsa/openmmv1x/trained_models/openeo/pretrain_BEN_10-20m_512-128/onnx' # for local debugging\n bands = ( 'B02', 'B03', 'B04', 'B05', 'B06', 'B07', 'B08', 'B8A', 'B11', 'B12')\n inspect(message=\"ULOGGING: model_dir set to default\", data=[model_dir, os.path.exists(model_dir)])\n print(f\"Path model: {model_dir}\")\n\n\n library_versions = {'python':sys.version, 'numpy': np.__version__, 'xarray': xarray.__version__, 'joblib': joblib.__version__, 'onnx_runtime': ort.__version__}\n library_versions = [f\"{k}: {v}\" for k,v in library_versions.items()]\n inspect(message=\"ULOGGING: library versions\", data=library_versions)\n\n\n ## get the array\n cubearray:xarray.DataArray = cube.get_array() # bands, y, x\n nan_mask = np.isnan(cubearray.data)\n nan_count = np.count_nonzero(nan_mask,axis=(1,2))\n\n inspect(message=\"ULOGGING: cubearray\", data=cubearray)\n inspect(message=\"ULOGGING: cubearray NaNs per band\", data= nan_count)\n\n print(f\"Cube array shape: {cubearray.shape}\")\n print(f\"Cube array NaNs: {np.count_nonzero(np.isnan(cubearray.data),axis=(1,2))}\")\n\n # replace NaNs with 0 so inference works\n if np.count_nonzero(nan_mask)>0:\n print(\"Replacing NaNs with 0\")\n cubearray = cubearray.interpolate_na(dim='y', method='linear', use_coordinate=False,limit=2 )/2 + cubearray.interpolate_na(dim='x', method='linear', use_coordinate=False, limit=2)/2\n cubearray = cubearray.fillna(0)\n \n # # Pixel size of the original image\n # init_pixel_size_x = cubearray.coords['x'][-1] - cubearray.coords['x'][-2]\n # init_pixel_size_y = cubearray.coords['y'][-1] - cubearray.coords['y'][-2]\n \n ## normalize data\n cubearray_normalized = preprocess_datacube(cubearray, scaler_dir=f'{model_dir}/scalers', ordered_bands=tuple(bands))\n\n ## inference\n result = process_window_onnx(cubearray_normalized, model_dir)\n # inspect(message=\"ULOGGING: result\", data=result)\n print(f\"Result shape: {result[0].shape}\")\n\n # coord_x = np.linspace(start=cube.get_array().coords['x'].min(), stop=cube.get_array().coords['x'].max() + init_pixel_size_x,\n # num=result.shape[-2], endpoint=False)\n # coord_y = np.linspace(start=cube.get_array().coords['y'].min(), stop=cube.get_array().coords['y'].max() + init_pixel_size_y,\n # num=result.shape[-1], endpoint=False)\n # print(cubearray.coords['x'].min())\n # print(cubearray.coords['y'].max())\n # print(init_pixel_size_x)\n \n # result_xarray = xarray.DataArray(result[0], dims=['bands', 'x', 'y'], coords=dict(x=coord_x, y=coord_y, bands=['level_0' ,'level_1']))\n\n ## transform numpy array predictions into an xarray with right dimension order for openEO XarrayDataCube\n result_xarray = xarray.DataArray(result[0], dims=[\"bands\", \"y\", \"x\"], coords=dict(bands=['level_0' ,'level_1'], x=cubearray.coords[\"x\"], y=cubearray.coords[\"y\"]))\n # inspect(message=\"ULOGGING: result_xarray\", data=result_xarray)\n return XarrayDataCube(result_xarray)\n \n except Exception as e:\n inspect(message=\"ULOGGING: Error in UDF!\",\n data=traceback.format_exc())\n exc_type, exc_obj, exc_tb = sys.exc_info()\n fname = os.path.split(exc_tb.tb_frame.f_code.co_filename)[1]\n print(exc_type, fname, exc_tb.tb_lineno)\n print(e)\n raise e\n" }, "process_id": "run_udf", "result": true } } }, "size": [ { "dimension": "x", "unit": "px", "value": 120 }, { "dimension": "y", "unit": "px", "value": 120 }, { "dimension": "t", "value": "P1D" } ] }, "process_id": "apply_neighborhood" }, "filterbands1": { "arguments": { "bands": [ "level_0", "level_1" ], "data": { "from_node": "renamelabels1" } }, "process_id": "filter_bands" }, "filterbands2": { "arguments": { "bands": [ "level_1" ], "data": { "from_node": "filterbands1" } }, "process_id": "filter_bands" }, "filterbands3": { "arguments": { "bands": [ "level_0" ], "data": { "from_node": "filterbands1" } }, "process_id": "filter_bands" }, "loadcollection1": { "arguments": { "bands": [ "B02", "B03", "B04", "B05", "B06", "B07", "B08", "B8A", "B11", "B12" ], "featureflags": { "tilesize": 120 }, "id": "SENTINEL2_L2A", "properties": { "eo:cloud_cover": { "process_graph": { "lte1": { "arguments": { "x": { "from_parameter": "value" }, "y": 10 }, "process_id": "lte", "result": true } } } }, "spatial_extent": { "from_parameter": "spatial_extent" }, "temporal_extent": { "from_parameter": "temporal_extent" } }, "process_id": "load_collection" }, "renamelabels1": { "arguments": { "data": { "from_node": "applyneighborhood1" }, "dimension": "bands", "target": [ "level_0", "level_1", "", "", "", "", "", "", "", "" ] }, "process_id": "rename_labels" }, "resamplespatial1": { "arguments": { "align": "upper-left", "data": { "from_node": "filterbands2" }, "method": "near", "projection": null, "resolution": 40 }, "process_id": "resample_spatial" }, "resamplespatial2": { "arguments": { "align": "upper-left", "data": { "from_node": "filterbands3" }, "method": "near", "projection": null, "resolution": 20 }, "process_id": "resample_spatial" }, "saveresult1": { "arguments": { "data": { "from_node": "resamplespatial1" }, "format": "GTiff", "options": { "filename_prefix": "level1_40m" } }, "process_id": "save_result" }, "saveresult2": { "arguments": { "data": { "from_node": "resamplespatial2" }, "format": "GTiff", "options": { "filename_prefix": "level0_20m" } }, "process_id": "save_result", "result": true } }, "public": true }