# Copyright 2019 The TensorFlow Authors. All Rights Reserved. # # 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. # ============================================================================== """TensorBoard 3D mesh visualizer plugin.""" import numpy as np from werkzeug import wrappers from tensorboard.backend import http_util from tensorboard.data import provider from tensorboard.plugins import base_plugin from tensorboard.plugins.mesh import metadata from tensorboard.plugins.mesh import plugin_data_pb2 from tensorboard import plugin_util _DEFAULT_DOWNSAMPLING = 100 # meshes per time series class MeshPlugin(base_plugin.TBPlugin): """A plugin that serves 3D visualization of meshes.""" plugin_name = metadata.PLUGIN_NAME def __init__(self, context): """Instantiates a MeshPlugin via TensorBoard core. Args: context: A base_plugin.TBContext instance. A magic container that TensorBoard uses to make objects available to the plugin. """ self._data_provider = context.data_provider self._downsample_to = (context.sampling_hints or {}).get( self.plugin_name, _DEFAULT_DOWNSAMPLING ) self._version_checker = plugin_util._MetadataVersionChecker( data_kind="mesh", latest_known_version=0, ) def _instance_tag_metadata(self, ctx, experiment, run, instance_tag): """Gets the `MeshPluginData` proto for an instance tag.""" results = self._data_provider.list_tensors( ctx, experiment_id=experiment, plugin_name=metadata.PLUGIN_NAME, run_tag_filter=provider.RunTagFilter( runs=[run], tags=[instance_tag] ), ) content = results[run][instance_tag].plugin_content return metadata.parse_plugin_metadata(content) def _tag(self, ctx, experiment, run, instance_tag): """Gets the user-facing tag name for an instance tag.""" return self._instance_tag_metadata( ctx, experiment, run, instance_tag ).name def _instance_tags(self, ctx, experiment, run, tag): """Gets the instance tag names for a user-facing tag.""" index = self._data_provider.list_tensors( ctx, experiment_id=experiment, plugin_name=metadata.PLUGIN_NAME, run_tag_filter=provider.RunTagFilter(runs=[run]), ) return [ instance_tag for (instance_tag, ts) in index.get(run, {}).items() if tag == metadata.parse_plugin_metadata(ts.plugin_content).name ] @wrappers.Request.application def _serve_tags(self, request): """A route (HTTP handler) that returns a response with tags. Args: request: The werkzeug.Request object. Returns: A response that contains a JSON object. The keys of the object are all the runs. Each run is mapped to a (potentially empty) list of all tags that are relevant to this plugin. """ ctx = plugin_util.context(request.environ) experiment = plugin_util.experiment_id(request.environ) all_runs = self._data_provider.list_tensors( ctx, experiment_id=experiment, plugin_name=metadata.PLUGIN_NAME, ) # tagToContent is itself a dictionary mapping tag name to string # SummaryMetadata.plugin_data.content. Retrieve the keys of that dictionary # to obtain a list of tags associated with each run. For each tag estimate # number of samples. response = dict() for run, tags in all_runs.items(): response[run] = dict() for (instance_tag, metadatum) in tags.items(): md = metadata.parse_plugin_metadata(metadatum.plugin_content) if not self._version_checker.ok(md.version, run, instance_tag): continue # Make sure we only operate on user-defined tags here. tag = self._tag(ctx, experiment, run, instance_tag) meta = self._instance_tag_metadata( ctx, experiment, run, instance_tag ) # Batch size must be defined, otherwise we don't know how many # samples were there. response[run][tag] = {"samples": meta.shape[0]} return http_util.Respond(request, response, "application/json") def get_plugin_apps(self): """Gets all routes offered by the plugin. This method is called by TensorBoard when retrieving all the routes offered by the plugin. Returns: A dictionary mapping URL path to route that handles it. """ # Note that the methods handling routes are decorated with # @wrappers.Request.application. return { "/tags": self._serve_tags, "/meshes": self._serve_mesh_metadata, "/data": self._serve_mesh_data, } def is_active(self): return False # `list_plugins` as called by TB core suffices def frontend_metadata(self): return base_plugin.FrontendMetadata(element_name="mesh-dashboard") def _get_sample(self, tensor_datum, sample): """Returns a single sample from a batch of samples.""" return tensor_datum.numpy[sample].tolist() def _get_tensor_metadata( self, event, content_type, components, data_shape, config ): """Converts a TensorDatum into a JSON-compatible response. Args: event: TensorDatum object containing data in proto format. content_type: enum plugin_data_pb2.MeshPluginData.ContentType value, representing content type in TensorDatum. components: Bitmask representing all parts (vertices, colors, etc.) that belong to the summary. data_shape: list of dimensions sizes of the tensor. config: rendering scene configuration as dictionary. Returns: Dictionary of transformed metadata. """ return { "wall_time": event.wall_time, "step": event.step, "content_type": content_type, "components": components, "config": config, "data_shape": list(data_shape), } def _get_tensor_data(self, event, sample): """Convert a TensorDatum into a JSON-compatible response.""" data = self._get_sample(event, sample) return data def _collect_tensor_events(self, request, step=None): """Collects list of tensor events based on request.""" ctx = plugin_util.context(request.environ) experiment = plugin_util.experiment_id(request.environ) run = request.args.get("run") tag = request.args.get("tag") tensor_events = [] # List of tuples (meta, tensor) that contain tag. for instance_tag in self._instance_tags(ctx, experiment, run, tag): tensors = self._data_provider.read_tensors( ctx, experiment_id=experiment, plugin_name=metadata.PLUGIN_NAME, run_tag_filter=provider.RunTagFilter( runs=[run], tags=[instance_tag] ), downsample=self._downsample_to, )[run][instance_tag] meta = self._instance_tag_metadata( ctx, experiment, run, instance_tag ) tensor_events += [(meta, tensor) for tensor in tensors] if step is not None: tensor_events = [ event for event in tensor_events if event[1].step == step ] else: # Make sure tensors sorted by step in ascending order. tensor_events = sorted( tensor_events, key=lambda tensor_data: tensor_data[1].step ) return tensor_events @wrappers.Request.application def _serve_mesh_data(self, request): """A route that returns data for particular summary of specified type. Data can represent vertices coordinates, vertices indices in faces, vertices colors and so on. Each mesh may have different combination of abovementioned data and each type/part of mesh summary must be served as separate roundtrip to the server. Args: request: werkzeug.Request containing content_type as a name of enum plugin_data_pb2.MeshPluginData.ContentType. Returns: werkzeug.Response either float32 or int32 data in binary format. """ step = float(request.args.get("step", 0.0)) tensor_events = self._collect_tensor_events(request, step) content_type = request.args.get("content_type") try: content_type = plugin_data_pb2.MeshPluginData.ContentType.Value( content_type ) except ValueError: return http_util.Respond( request, "Bad content_type", "text/plain", 400 ) sample = int(request.args.get("sample", 0)) response = [ self._get_tensor_data(tensor, sample) for meta, tensor in tensor_events if meta.content_type == content_type ] np_type = { plugin_data_pb2.MeshPluginData.VERTEX: np.float32, plugin_data_pb2.MeshPluginData.FACE: np.int32, plugin_data_pb2.MeshPluginData.COLOR: np.uint8, }[content_type] response = np.array(response, dtype=np_type) # Looks like reshape can take around 160ms, so why not store it reshaped. response = response.reshape(-1).tobytes() return http_util.Respond(request, response, "arraybuffer") @wrappers.Request.application def _serve_mesh_metadata(self, request): """A route that returns the mesh metadata associated with a tag. Metadata consists of wall time, type of elements in tensor, scene configuration and so on. Args: request: The werkzeug.Request object. Returns: A JSON list of mesh data associated with the run and tag combination. """ tensor_events = self._collect_tensor_events(request) # We convert the tensor data to text. response = [ self._get_tensor_metadata( tensor, meta.content_type, meta.components, meta.shape, meta.json_config, ) for meta, tensor in tensor_events ] return http_util.Respond(request, response, "application/json")