megadetector 5.0.10__py3-none-any.whl → 5.0.12__py3-none-any.whl

api/synchronous/api_core/animal_detection_api/detection/tf_detector.py

@@ -1,165 +0,0 @@
-########
-#
-# tf_detector.py
-#
-# Module containing the class TFDetector for loading a TensorFlow detection model and
-# running inference.
-# 
-########
-
-import numpy as np
-
-from detection.run_detector import CONF_DIGITS, COORD_DIGITS, FAILURE_INFER
-from md_utils.ct_utils import truncate_float
-
-import tensorflow.compat.v1 as tf
-
-print('TensorFlow version:', tf.__version__)
-print('Is GPU available? tf.test.is_gpu_available:', tf.test.is_gpu_available())
-
-
-class TFDetector:
-    """
-    A detector model loaded at the time of initialization. It is intended to be used with
-    the MegaDetector (TF). The inference batch size is set to 1; code needs to be modified
-    to support larger batch sizes, including resizing appropriately.
-    """
-    
-    # MegaDetector was trained with batch size of 1, and the resizing function is a part
-    # of the inference graph
-    BATCH_SIZE = 1
-
-
-    def __init__(self, model_path):
-        """
-        Loads model from model_path and starts a tf.Session with this graph. Obtains
-        input and output tensor handles.
-        """
-        
-        detection_graph = TFDetector.__load_model(model_path)
-        self.tf_session = tf.Session(graph=detection_graph)
-
-        self.image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
-        self.box_tensor = detection_graph.get_tensor_by_name('detection_boxes:0')
-        self.score_tensor = detection_graph.get_tensor_by_name('detection_scores:0')
-        self.class_tensor = detection_graph.get_tensor_by_name('detection_classes:0')
-
-    @staticmethod
-    def round_and_make_float(d, precision=4):
-        return truncate_float(float(d), precision=precision)
-
-    @staticmethod
-    def __convert_coords(tf_coords):
-        """
-        Converts coordinates from the model's output format [y1, x1, y2, x2] to the
-        format used by our API and MegaDB: [x1, y1, width, height]. All coordinates
-        (including model outputs) are normalized in the range [0, 1].
-
-        Args:
-            tf_coords: np.array of predicted bounding box coordinates from the TF detector,
-                has format [y1, x1, y2, x2]
-
-        Returns: list of Python float, predicted bounding box coordinates [x1, y1, width, height]
-        """
-        
-        # change from [y1, x1, y2, x2] to [x1, y1, width, height]
-        width = tf_coords[3] - tf_coords[1]
-        height = tf_coords[2] - tf_coords[0]
-
-        new = [tf_coords[1], tf_coords[0], width, height]  # must be a list instead of np.array
-
-        # convert numpy floats to Python floats
-        for i, d in enumerate(new):
-            new[i] = TFDetector.round_and_make_float(d, precision=COORD_DIGITS)
-        return new
-
-    @staticmethod
-    def __load_model(model_path):
-        """
-        Loads a detection model (i.e., create a graph) from a .pb file.
-
-        Args:
-            model_path: .pb file of the model.
-
-        Returns: the loaded graph.
-        """
-        
-        print('TFDetector: Loading graph...')
-        detection_graph = tf.Graph()
-        with detection_graph.as_default():
-            od_graph_def = tf.GraphDef()
-            with tf.gfile.GFile(model_path, 'rb') as fid:
-                serialized_graph = fid.read()
-                od_graph_def.ParseFromString(serialized_graph)
-                tf.import_graph_def(od_graph_def, name='')
-        print('TFDetector: Detection graph loaded.')
-
-        return detection_graph
-
-    def _generate_detections_one_image(self, image):
-        np_im = np.asarray(image, np.uint8)
-        im_w_batch_dim = np.expand_dims(np_im, axis=0)
-
-        # need to change the above line to the following if supporting a batch size > 1 and resizing to the same size
-        # np_images = [np.asarray(image, np.uint8) for image in images]
-        # images_stacked = np.stack(np_images, axis=0) if len(images) > 1 else np.expand_dims(np_images[0], axis=0)
-
-        # performs inference
-        (box_tensor_out, score_tensor_out, class_tensor_out) = self.tf_session.run(
-            [self.box_tensor, self.score_tensor, self.class_tensor],
-            feed_dict={self.image_tensor: im_w_batch_dim})
-
-        return box_tensor_out, score_tensor_out, class_tensor_out
-
-    def generate_detections_one_image(self, image, image_id, detection_threshold, image_size=None,
-                                      skip_image_resizing=False):
-        """
-        Apply the detector to an image.
-
-        Args:
-            image: the PIL Image object
-            image_id: a path to identify the image; will be in the "file" field of the output object
-            detection_threshold: confidence above which to include the detection proposal
-
-        Returns:
-        A dict with the following fields, see the 'images' key in https://github.com/agentmorris/MegaDetector/tree/master/api/batch_processing#batch-processing-api-output-format
-            - 'file' (always present)
-            - 'max_detection_conf'
-            - 'detections', which is a list of detection objects containing keys 'category', 'conf' and 'bbox'
-            - 'failure'
-        """
-        
-        assert image_size is None, 'Image sizing not supported for TF detectors'
-        assert not skip_image_resizing, 'Image sizing not supported for TF detectors'
-        result = {
-            'file': image_id
-        }
-        try:
-            b_box, b_score, b_class = self._generate_detections_one_image(image)
-
-            # our batch size is 1; need to loop the batch dim if supporting batch size > 1
-            boxes, scores, classes = b_box[0], b_score[0], b_class[0]
-
-            detections_cur_image = []  # will be empty for an image with no confident detections
-            max_detection_conf = 0.0
-            for b, s, c in zip(boxes, scores, classes):
-                if s > detection_threshold:
-                    detection_entry = {
-                        'category': str(int(c)),  # use string type for the numerical class label, not int
-                        'conf': truncate_float(float(s),  # cast to float for json serialization
-                                               precision=CONF_DIGITS),
-                        'bbox': TFDetector.__convert_coords(b)
-                    }
-                    detections_cur_image.append(detection_entry)
-                    if s > max_detection_conf:
-                        max_detection_conf = s
-
-            result['max_detection_conf'] = truncate_float(float(max_detection_conf),
-                                                          precision=CONF_DIGITS)
-            result['detections'] = detections_cur_image
-
-        except Exception as e:
-            result['failure'] = FAILURE_INFER
-            print('TFDetector: image {} failed during inference: {}'.format(image_id, str(e)))
-
-        return result

api/synchronous/api_core/animal_detection_api/detection/video_utils.py

@@ -1,495 +0,0 @@
-########
-#
-# video_utils.py
-#
-# Utilities for splitting, rendering, and assembling videos.
-#
-########
-
-#%% Constants, imports, environment
-
-import os
-import cv2
-import glob
-import json
-
-from collections import defaultdict
-from multiprocessing.pool import ThreadPool
-from multiprocessing.pool import Pool
-from tqdm import tqdm
-from typing import Container,Iterable,List
-from functools import partial
-
-from md_utils import path_utils
-    
-from md_visualization import visualization_utils as vis_utils
-
-
-#%% Path utilities
-
-VIDEO_EXTENSIONS = ('.mp4','.avi','.mpeg','.mpg')
-
-def is_video_file(s: str, video_extensions: Container[str] = VIDEO_EXTENSIONS
-                  ) -> bool:
-    """
-    Checks a file's extension against a hard-coded set of video file
-    extensions.
-    """
-    
-    ext = os.path.splitext(s)[1]
-    return ext.lower() in video_extensions
-
-
-def find_video_strings(strings: Iterable[str]) -> List[str]:
-    """
-    Given a list of strings that are potentially video file names, looks for
-    strings that actually look like video file names (based on extension).
-    """
-    
-    return [s for s in strings if is_video_file(s.lower())]
-
-
-def find_videos(dirname: str, recursive: bool = False) -> List[str]:
-    """
-    Finds all files in a directory that look like video file names. Returns
-    absolute paths.
-    """
-    
-    if recursive:
-        strings = glob.glob(os.path.join(dirname, '**', '*.*'), recursive=True)
-    else:
-        strings = glob.glob(os.path.join(dirname, '*.*'))
-    return find_video_strings(strings)
-
-
-#%% Function for rendering frames to video and vice-versa
-
-# http://tsaith.github.io/combine-images-into-a-video-with-python-3-and-opencv-3.html
-
-def frames_to_video(images, Fs, output_file_name, codec_spec='h264'):
-    """
-    Given a list of image files and a sample rate, concatenate those images into
-    a video and write to [output_file_name].
-    
-    Note to self: h264 is a sensible default and generally works on Windows, but when this
-    fails (which is around 50% of the time on Linux), I fall back to mp4v.
-    """
-    
-    if codec_spec is None:
-        codec_spec = 'h264'
-        
-    if len(images) == 0:
-        return
-
-    # Determine the width and height from the first image
-    frame = cv2.imread(images[0])
-    cv2.imshow('video',frame)
-    height, width, channels = frame.shape
-
-    # Define the codec and create VideoWriter object
-    fourcc = cv2.VideoWriter_fourcc(*codec_spec)
-    out = cv2.VideoWriter(output_file_name, fourcc, Fs, (width, height))
-
-    for image in images:
-        frame = cv2.imread(image)
-        out.write(frame)
-
-    out.release()
-    cv2.destroyAllWindows()
-
-
-def get_video_fs(input_video_file):
-    """
-    Get the frame rate of [input_video_file]
-    """
-    
-    assert os.path.isfile(input_video_file), 'File {} not found'.format(input_video_file)    
-    vidcap = cv2.VideoCapture(input_video_file)
-    Fs = vidcap.get(cv2.CAP_PROP_FPS)
-    vidcap.release()
-    return Fs
-
-
-def frame_number_to_filename(frame_number):
-    return 'frame{:06d}.jpg'.format(frame_number)
-
-
-def video_to_frames(input_video_file, output_folder, overwrite=True, 
-                    every_n_frames=None, verbose=False):
-    """
-    Render every frame of [input_video_file] to a .jpg in [output_folder]
-    
-    With help from:
-        
-    https://stackoverflow.com/questions/33311153/python-extracting-and-saving-video-frames
-    """
-    
-    assert os.path.isfile(input_video_file), 'File {} not found'.format(input_video_file)
-    
-    vidcap = cv2.VideoCapture(input_video_file)
-    n_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
-    Fs = vidcap.get(cv2.CAP_PROP_FPS)
-    
-    # If we're not over-writing, check whether all frame images already exist
-    if overwrite == False:
-        
-        missing_frame_number = None
-        frame_filenames = []
-        
-        for frame_number in range(0,n_frames):
-            
-            if every_n_frames is not None:
-                if frame_number % every_n_frames != 0:
-                    continue
-            
-            frame_filename = frame_number_to_filename(frame_number)
-            frame_filename = os.path.join(output_folder,frame_filename)
-            frame_filenames.append(frame_filename)
-            if os.path.isfile(frame_filename):
-                continue
-            else:
-                missing_frame_number = frame_number
-                break
-    
-        # OpenCV seems to over-report the number of frames by 1 in some cases, or fails
-        # to read the last frame; either way, I'm allowing one missing frame.
-        allow_last_frame_missing = True
-        
-        if missing_frame_number is None or \
-            (allow_last_frame_missing and (missing_frame_number == n_frames-1)):
-            if verbose:
-                print('Skipping video {}, all output frames exist'.format(input_video_file))
-            return frame_filenames,Fs
-        else:
-            pass
-            # print("Rendering video {}, couldn't find frame {}".format(
-            #    input_video_file,missing_frame_number))
-    
-    # ...if we need to check whether to skip this video entirely
-        
-    if verbose:
-        print('Reading {} frames at {} Hz from {}'.format(n_frames,Fs,input_video_file))
-
-    frame_filenames = []
-
-    # for frame_number in tqdm(range(0,n_frames)):
-    for frame_number in range(0,n_frames):
-
-        success,image = vidcap.read()
-        if not success:
-            assert image is None
-            if verbose:
-                print('Read terminating at frame {} of {}'.format(frame_number,n_frames))
-            break
-
-        if every_n_frames is not None:
-            if frame_number % every_n_frames != 0:
-                continue
-            
-        frame_filename = frame_number_to_filename(frame_number)
-        frame_filename = os.path.join(output_folder,frame_filename)
-        frame_filenames.append(frame_filename)
-        
-        if overwrite == False and os.path.isfile(frame_filename):
-            # print('Skipping frame {}'.format(frame_filename))
-            pass            
-        else:
-            try:
-                if frame_filename.isascii():
-                    cv2.imwrite(os.path.normpath(frame_filename),image)
-                else:
-                    is_success, im_buf_arr = cv2.imencode('.jpg', image)
-                    im_buf_arr.tofile(frame_filename)
-                assert os.path.isfile(frame_filename), \
-                    'Output frame {} unavailable'.format(frame_filename)
-            except KeyboardInterrupt:
-                vidcap.release()
-                raise
-            except Exception as e:
-                print('Error on frame {} of {}: {}'.format(frame_number,n_frames,str(e)))
-
-    if verbose:
-        print('\nExtracted {} of {} frames'.format(len(frame_filenames),n_frames))
-
-    vidcap.release()    
-    return frame_filenames,Fs
-
-
-def _video_to_frames_for_folder(relative_fn,input_folder,output_folder_base,every_n_frames,overwrite,verbose):
-    """
-    Internal function to call video_to_frames in the context of video_folder_to_frames; 
-    makes sure the right output folder exists, then calls video_to_frames.
-    """    
-    
-    input_fn_absolute = os.path.join(input_folder,relative_fn)
-    assert os.path.isfile(input_fn_absolute)
-
-    # Create the target output folder
-    output_folder_video = os.path.join(output_folder_base,relative_fn)
-    os.makedirs(output_folder_video,exist_ok=True)
-
-    # Render frames
-    # input_video_file = input_fn_absolute; output_folder = output_folder_video
-    frame_filenames,fs = video_to_frames(input_fn_absolute,output_folder_video,
-                                         overwrite=overwrite,every_n_frames=every_n_frames,
-                                         verbose=verbose)
-    
-    return frame_filenames,fs
-
-
-def video_folder_to_frames(input_folder:str, output_folder_base:str, 
-                           recursive:bool=True, overwrite:bool=True,
-                           n_threads:int=1, every_n_frames:int=None,
-                           verbose=False, parallelization_uses_threads=True):
-    """
-    For every video file in input_folder, create a folder within output_folder_base, and 
-    render every frame of the video to .jpg in that folder.
-    """
-    
-    # Recursively enumerate video files
-    input_files_full_paths = find_videos(input_folder,recursive=recursive)
-    print('Found {} videos in folder {}'.format(len(input_files_full_paths),input_folder))
-    if len(input_files_full_paths) == 0:
-        return [],[],[]
-    
-    input_files_relative_paths = [os.path.relpath(s,input_folder) for s in input_files_full_paths]
-    input_files_relative_paths = [s.replace('\\','/') for s in input_files_relative_paths]
-    
-    os.makedirs(output_folder_base,exist_ok=True)    
-    
-    frame_filenames_by_video = []
-    fs_by_video = []
-    
-    if n_threads == 1:
-        # For each video
-        #
-        # input_fn_relative = input_files_relative_paths[0]
-        for input_fn_relative in tqdm(input_files_relative_paths):
-        
-            frame_filenames,fs = \
-                _video_to_frames_for_folder(input_fn_relative,input_folder,output_folder_base,
-                                            every_n_frames,overwrite,verbose)
-            frame_filenames_by_video.append(frame_filenames)
-            fs_by_video.append(fs)
-    else:
-        if parallelization_uses_threads:
-            print('Starting a worker pool with {} threads'.format(n_threads))
-            pool = ThreadPool(n_threads)
-        else:
-            print('Starting a worker pool with {} processes'.format(n_threads))
-            pool = Pool(n_threads)
-        process_video_with_options = partial(_video_to_frames_for_folder, 
-                                             input_folder=input_folder,
-                                             output_folder_base=output_folder_base,
-                                             every_n_frames=every_n_frames,
-                                             overwrite=overwrite,
-                                             verbose=verbose)
-        results = list(tqdm(pool.imap(
-            partial(process_video_with_options),input_files_relative_paths), 
-                            total=len(input_files_relative_paths)))
-        frame_filenames_by_video = [x[0] for x in results]
-        fs_by_video = [x[1] for x in results]
-        
-    return frame_filenames_by_video,fs_by_video,input_files_full_paths
-  
-
-class FrameToVideoOptions:
-    
-    # zero-indexed
-    nth_highest_confidence = 1
-    
-    
-def frame_results_to_video_results(input_file,output_file,options:FrameToVideoOptions = None):
-    """
-    Given an API output file produced at the *frame* level, corresponding to a directory 
-    created with video_folder_to_frames, map those frame-level results back to the 
-    video level for use in Timelapse.
-    
-    Preserves everything in the input .json file other than the images.
-    """
-
-    if options is None:
-        options = FrameToVideoOptions()
-        
-    # Load results
-    with open(input_file,'r') as f:
-        input_data = json.load(f)
-
-    images = input_data['images']
-    detection_categories = input_data['detection_categories']
-    
-    ## Break into videos
-    
-    video_to_frames = defaultdict(list) 
-    
-    # im = images[0]
-    for im in tqdm(images):
-        
-        fn = im['file']
-        video_name = os.path.dirname(fn)
-        assert is_video_file(video_name)
-        video_to_frames[video_name].append(im)
-    
-    print('Found {} unique videos in {} frame-level results'.format(
-        len(video_to_frames),len(images)))
-    
-    output_images = []
-    
-    ## For each video...
-    
-    # video_name = list(video_to_frames.keys())[0]
-    for video_name in tqdm(video_to_frames):
-        
-        frames = video_to_frames[video_name]
-        
-        all_detections_this_video = []
-        
-        # frame = frames[0]
-        for frame in frames:
-            if frame['detections'] is not None:
-                all_detections_this_video.extend(frame['detections'])
-            
-        # At most one detection for each category for the whole video
-        canonical_detections = []
-            
-        # category_id = list(detection_categories.keys())[0]
-        for category_id in detection_categories:
-            
-            category_detections = [det for det in all_detections_this_video if \
-                                   det['category'] == category_id]
-            
-            # Find the nth-highest-confidence video to choose a confidence value
-            if len(category_detections) >= options.nth_highest_confidence:
-                
-                category_detections_by_confidence = sorted(category_detections, 
-                                                           key = lambda i: i['conf'],reverse=True)
-                canonical_detection = category_detections_by_confidence[options.nth_highest_confidence-1]
-                canonical_detections.append(canonical_detection)
-                                      
-        # Prepare the output representation for this video
-        im_out = {}
-        im_out['file'] = video_name
-        im_out['detections'] = canonical_detections
-        
-        # 'max_detection_conf' is no longer included in output files by default
-        if False:
-            im_out['max_detection_conf'] = 0
-            if len(canonical_detections) > 0:
-                confidences = [d['conf'] for d in canonical_detections]
-                im_out['max_detection_conf'] = max(confidences)
-        
-        output_images.append(im_out)
-        
-    # ...for each video
-    
-    output_data = input_data
-    output_data['images'] = output_images
-    s = json.dumps(output_data,indent=1)
-    
-    # Write the output file
-    with open(output_file,'w') as f:
-        f.write(s)
-    
-        
-#%% Test driver
-
-if False:
-
-    #%% Constants
-    
-    Fs = 30.01
-    confidence_threshold = 0.75
-    input_folder = 'z:\\'
-    frame_folder_base = r'e:\video_test\frames'
-    detected_frame_folder_base = r'e:\video_test\detected_frames'
-    rendered_videos_folder_base = r'e:\video_test\rendered_videos'
-    
-    results_file = r'results.json'
-    os.makedirs(detected_frame_folder_base,exist_ok=True)
-    os.makedirs(rendered_videos_folder_base,exist_ok=True)
-    
-    
-    #%% Split videos into frames
-        
-    frame_filenames_by_video,fs_by_video,video_filenames = \
-        video_folder_to_frames(input_folder,frame_folder_base,recursive=True)
-    
-    
-    #%% List image files, break into folders
-    
-    frame_files = path_utils.find_images(frame_folder_base,True)
-    frame_files = [s.replace('\\','/') for s in frame_files]
-    print('Enumerated {} total frames'.format(len(frame_files)))
-    
-    Fs = 30.01
-    # Find unique folders
-    folders = set()
-    # fn = frame_files[0]
-    for fn in frame_files:
-        folders.add(os.path.dirname(fn))
-    folders = [s.replace('\\','/') for s in folders]
-    print('Found {} folders for {} files'.format(len(folders),len(frame_files)))
-    
-        
-    #%% Load detector output
-    
-    with open(results_file,'r') as f:
-        detection_results = json.load(f)
-    detections = detection_results['images']
-    detector_label_map = detection_results['detection_categories']
-    for d in detections:
-        d['file'] = d['file'].replace('\\','/').replace('video_frames/','')
-
-
-    #%% Render detector frames
-    
-    # folder = list(folders)[0]
-    for folder in folders:
-        
-        frame_files_this_folder = [fn for fn in frame_files if folder in fn]
-        folder_relative = folder.replace((frame_folder_base + '/').replace('\\','/'),'')
-        detection_results_this_folder = [d for d in detections if folder_relative in d['file']]
-        print('Found {} detections in folder {}'.format(len(detection_results_this_folder),folder))
-        assert len(frame_files_this_folder) == len(detection_results_this_folder)
-        
-        rendered_frame_output_folder = os.path.join(detected_frame_folder_base,folder_relative)
-        os.makedirs(rendered_frame_output_folder,exist_ok=True)
-        
-        # d = detection_results_this_folder[0]
-        for d in tqdm(detection_results_this_folder):
-            
-            input_file = os.path.join(frame_folder_base,d['file'])
-            output_file = os.path.join(detected_frame_folder_base,d['file'])
-            os.makedirs(os.path.dirname(output_file),exist_ok=True)
-            vis_utils.draw_bounding_boxes_on_file(input_file,output_file,d['detections'],
-                                                  confidence_threshold)
-        
-        # ...for each file in this folder
-            
-    # ...for each folder
-
-
-    #%% Render output videos
-            
-    # folder = list(folders)[0]
-    for folder in tqdm(folders):
-        
-        folder_relative = folder.replace((frame_folder_base + '/').replace('\\','/'),'')
-        rendered_detector_output_folder = os.path.join(detected_frame_folder_base,folder_relative)
-        assert os.path.isdir(rendered_detector_output_folder)
-        
-        frame_files_relative = os.listdir(rendered_detector_output_folder)
-        frame_files_absolute = [os.path.join(rendered_detector_output_folder,s) \
-                                for s in frame_files_relative]
-        
-        output_video_filename = os.path.join(rendered_videos_folder_base,folder_relative)
-        os.makedirs(os.path.dirname(output_video_filename),exist_ok=True)
-        
-        original_video_filename = output_video_filename.replace(
-            rendered_videos_folder_base,input_folder)
-        assert os.path.isfile(original_video_filename)
-        Fs = get_video_fs(original_video_filename)
-                
-        frames_to_video(frame_files_absolute, Fs, output_video_filename)
-
-    # ...for each video