megadetector 5.0.21__py3-none-any.whl → 5.0.23__py3-none-any.whl

megadetector/data_management/importers/raic_csv_to_md_results.py

@@ -0,0 +1,416 @@
+"""
+
+raic_csv_to_md_results.py
+
+Converts classification+detection results in the .csv format provided to the Snapshot
+Serengeti program by the RAIC team to the MD results format.
+
+The input format is two .csv files:
+    
+* One with results, with columns [unnamed], filename, category, x_center, y_center,
+  width, height, confidence, datetime
+
+* One with class IDs and names, with columns CLASS, SPECIES
+
+Filenames are relative paths to .txt files, but with slashes replaced by underscores, e.g. this 
+file:
+    
+    B04_R1/I__00122.JPG
+    
+...appears in the .csv file as:
+    
+    B04_R1_I__00122.txt
+    
+Image coordinates are in absolute floating-point units, with an upper-left origin.
+  
+Unknowns at the time I'm writing this:
+    
+* I don't know what the unnamed column is, but it looks like an ID I can safely ignore.
+
+* I believe that MegaDetector was run, then a classifier was run, but there is a 
+  single "confidence" column in the output.  I am writing out the results as if they were a 
+  single multi-class detector.  This is suspicious given the lack of a human class, which suggests
+  that this is intended to be run in conjunection with MD.
+
+* There is no concept of "empty" in this file format, so by default I assume that images with
+  no annotations in the .csv file were processed and determine to have no detections above some
+  (unknown) threshold.
+  
+* I'm not currently handling EXIF rotations, as part of the effort to simplify this file
+  for conversion to R (see below).
+  
+Note to self: this file should not take dependencies on other components of the MD
+repo, at the risk of creating some redundancy.  I am going to convert this to R,
+which will be easier if it's not using any non-standard libraries.  Anything in the 
+"interactive driver" cells gets a pass.
+
+"""
+
+#%% Imports and constants
+
+import os
+import glob
+import json
+import sys
+import argparse
+
+import pandas as pd
+from PIL import Image
+
+
+#%% Functions from the MD python package
+
+# ...that I'm choosing to copy and paste to facilitate a conversion of this
+# script to R.
+
+# Should all be lower-case
+IMG_EXTENSIONS = ('.jpg', '.jpeg', '.gif', '.png', '.tif', '.tiff', '.bmp')
+
+def _is_image_file(s, img_extensions=IMG_EXTENSIONS):
+    """
+    Checks a file's extension against a hard-coded set of image file
+    extensions.  Uses case-insensitive comparison.
+    
+    Does not check whether the file exists, only determines whether the filename
+    implies it's an image file.
+    
+    Args:
+        s (str): filename to evaluate for image-ness
+        img_extensions (list, optional): list of known image file extensions
+        
+    Returns:
+        bool: True if [s] appears to be an image file, else False
+    """
+    
+    ext = os.path.splitext(s)[1]
+    return ext.lower() in img_extensions
+
+
+def _find_image_strings(strings):
+    """
+    Given a list of strings that are potentially image file names, looks for
+    strings that actually look like image file names (based on extension).
+    
+    Args:
+        strings (list): list of filenames to check for image-ness
+        
+    Returns:
+        list: the subset of [strings] that appear to be image filenames
+    """
+    
+    return [s for s in strings if _is_image_file(s)]
+
+
+def _find_images(dirname, 
+                recursive=False, 
+                return_relative_paths=False, 
+                convert_slashes=True):
+    """
+    Finds all files in a directory that look like image file names. Returns
+    absolute paths unless return_relative_paths is set.  Uses the OS-native
+    path separator unless convert_slashes is set, in which case will always
+    use '/'.
+    
+    Args:
+        dirname (str): the folder to search for images
+        recursive (bool, optional): whether to search recursively
+        return_relative_paths (str, optional): return paths that are relative
+            to [dirname], rather than absolute paths
+        convert_slashes (bool, optional): force forward slashes in return values
+
+    Returns:
+        list: list of image filenames found in [dirname]
+    """
+    
+    assert os.path.isdir(dirname), '{} is not a folder'.format(dirname)
+    
+    if recursive:
+        strings = glob.glob(os.path.join(dirname, '**', '*.*'), recursive=True)
+    else:
+        strings = glob.glob(os.path.join(dirname, '*.*'))
+    
+    image_files = _find_image_strings(strings)
+    
+    if return_relative_paths:
+        image_files = [os.path.relpath(fn,dirname) for fn in image_files]
+    
+    image_files = sorted(image_files)
+    
+    if convert_slashes:
+        image_files = [fn.replace('\\', '/') for fn in image_files]
+        
+    return image_files
+
+
+#%% Main conversion function
+
+def raic_csv_to_md_results(result_csv_file,
+                           class_mapping_csv_file,
+                           image_folder,
+                           output_file=None,
+                           unannotated_image_handling='empty'):
+    """
+    Converts a pair of .csv files (see file header for details) to MD results format.
+    
+    Currently errors if image filenames are ambiguous, or if any images referred to in 
+    the results are not available.
+    
+    Args:
+        result_csv_file (str): the results file to read (.csv)
+        class_mapping_csv_file (str): the class mapping file (.csv)
+        image_folder (str): the folder containing all the images referred to in
+            [result_csv_file]
+        output_file (str, optional): the .json file to which we should write results.  Defaults
+            to [result_csv_file].json
+        unannotated_image_handling (str, optional): can be "empty" (default) to assume
+            that images without annotations are empty, "warning", "error", or "skip"
+            
+    Returns:
+        str: the output file written, identical to [output_file] if [output_file] was not None
+    """
+    
+    # Validate arguments    
+    assert os.path.isfile(result_csv_file), \
+        'Result file {} not found'.format(result_csv_file)
+    assert os.path.isfile(class_mapping_csv_file), \
+        'Class mapping file {} not found'.format(class_mapping_csv_file)
+    assert os.path.isdir(image_folder), \
+        'Image folder {} not found'.format(image_folder)
+    
+    if output_file is None:
+        output_file = result_csv_file + '.json'
+        
+    image_files_relative = _find_images(image_folder,
+                                        recursive=True,
+                                        return_relative_paths=True,
+                                        convert_slashes=True)
+    image_file_base_flattened_to_image_file_relative = {}
+    for fn in image_files_relative:
+        # Convert, e.g. B04_R1/I__00108.JPG to B04_R1_I__00108
+        fn_flattened = fn.replace('/','_')
+        fn_flattened_base = os.path.splitext(fn_flattened)[0]
+        image_file_base_flattened_to_image_file_relative[fn_flattened_base] = \
+            fn    
+        
+    # Read the .csv files
+    df_results = pd.read_csv(result_csv_file)
+    df_class_mapping = pd.read_csv(class_mapping_csv_file)
+    
+    assert 'CLASS' in df_class_mapping.columns and 'SPECIES' in df_class_mapping.columns, \
+        'Unexpected column names in class mapping file {}'.format(class_mapping_csv_file)
+    
+    category_id_to_name = {}
+    for i_row,row in df_class_mapping.iterrows():
+        class_id = int(row['CLASS'])
+        assert class_id not in category_id_to_name, \
+            'Class ID {} occurs more than once in class mapping file {}'.format(
+                class_id,class_mapping_csv_file)
+        category_id_to_name[class_id] = row['SPECIES']
+    
+    if len(category_id_to_name) != len(set(category_id_to_name.values())):
+        print('Warning: one or more categories are used more than once in class mapping file {}'.format(
+            class_mapping_csv_file))
+        
+    # Convert results
+    
+    fn_relative_to_im = {}
+    
+    # i_row = 0; row = df_results.iloc[i_row]
+    for i_row,row in df_results.iterrows():
+        
+        # Map the .txt filename base to a relative path
+        bn = row['filename']
+        assert bn.lower().endswith('.txt')
+        bn_no_ext = os.path.splitext(bn)[0]
+        assert bn_no_ext in image_file_base_flattened_to_image_file_relative, \
+            'No image found for result row {}'.format(row['filename'])
+        
+        image_fn_relative = image_file_base_flattened_to_image_file_relative[bn_no_ext]
+        
+        # Have we seen another detection for this image?
+        if image_fn_relative in fn_relative_to_im:
+            
+            im = fn_relative_to_im[image_fn_relative]
+        
+        # If not, load this image so we can read its size
+        else:
+            
+            image_fn_abs = os.path.join(image_folder,image_fn_relative)
+            image = Image.open(image_fn_abs)
+            w = image.size[0]
+            h = image.size[1]
+            
+            im = {}
+            im['file'] = image_fn_relative
+            im['width'] = w
+            im['height'] = h
+            im['detections'] = []
+            im['datetime'] = str(row['datetime'])
+            fn_relative_to_im[image_fn_relative] = im
+            
+        # Convert annotation
+        x_center_abs = row['x_center']
+        y_center_abs = row['y_center']
+        box_width_abs = row['width']
+        box_height_abs = row['height']
+        
+        # Convert to relative coordinates
+        box_left_abs = x_center_abs - (box_width_abs/2.0)
+        box_top_abs = y_center_abs - (box_height_abs/2.0)
+        bbox_normalized = [box_left_abs/im['width'],
+                           box_top_abs/im['height'],
+                           box_width_abs/im['width'],
+                           box_height_abs/im['height']]
+        
+        category_id = str(int(row['category']))
+        confidence = row['confidence']        
+        assert isinstance(confidence,float) and confidence <= 1.0 and confidence >= 0.0
+        
+        det = {}
+        im['detections'].append(det)
+        det['category'] = category_id
+        det['conf'] = confidence
+        det['bbox'] = bbox_normalized
+                
+    # ...for each row
+        
+    n_empty_images = 0
+    
+    # Handle images without annotations
+    for fn_relative in image_files_relative:
+        
+        if fn_relative not in fn_relative_to_im:
+            if unannotated_image_handling == 'empty':
+                im = {}
+                im['file'] = fn_relative
+                im['detections'] = []                
+                fn_relative_to_im[fn_relative] = im
+                n_empty_images += 1
+                # Don't bother to read width and height here
+            elif unannotated_image_handling == 'warning':
+                print('Warning: image {} is not represented in the .csv results file'.format(fn_relative))
+            elif unannotated_image_handling == 'error':
+                raise ValueError('Image {} is not represented in the .csv results file'.format(fn_relative))
+            elif unannotated_image_handling == 'skip':
+                continue
+            
+    # ...for each image file
+    
+    if n_empty_images > 0:
+        print('Warning: assuming {} of {} images without annotations are empty'.format(
+            n_empty_images,len(image_files_relative)))
+        
+    images = list(fn_relative_to_im.values())
+            
+    # The MD output format uses string-ints for category IDs, right now we have ints
+    detection_categories = {}
+    for category_id_int in category_id_to_name:
+        detection_categories[str(category_id_int)] = category_id_to_name[category_id_int]
+    
+    info = {}
+    info['format_version'] = '1.4'
+    info['detector'] = 'RAIC .csv converter'
+    
+    d = {}
+    d['images'] = images
+    d['detection_categories'] = detection_categories
+    d['info'] = info
+        
+    with open(output_file,'w') as f:
+        json.dump(d,f,indent=1)
+        
+    return output_file
+
+# ...def raic_csv_to_md_results(...)
+
+    
+#%% Interactive driver
+
+if False:
+    
+    pass
+
+    #%% Test conversion
+    
+    base_folder = r'G:\temp\S24_B04_R1_output_annotations_for_Dan'
+    result_csv_file = os.path.join(base_folder,'S24_B04_R1_output_annotations_for_Dan.csv')
+    class_mapping_csv_file = os.path.join(base_folder,'categories_key.csv')
+    
+    # This is wrong, B04_R1 has to be part of the image paths
+    # image_folder = os.path.join(base_folder,'B04_R1')
+    
+    image_folder = base_folder 
+    
+    output_file = None
+    unannotated_image_handling='empty'
+    
+    output_file = raic_csv_to_md_results(result_csv_file=result_csv_file, 
+                                         class_mapping_csv_file=class_mapping_csv_file,
+                                         image_folder=image_folder,
+                                         output_file=output_file,
+                                         unannotated_image_handling=unannotated_image_handling)
+
+    #%% Validate results file
+    
+    from megadetector.postprocessing.validate_batch_results import \
+        ValidateBatchResultsOptions, validate_batch_results
+        
+    validation_options = ValidateBatchResultsOptions()
+    validation_options.check_image_existence = True
+    validation_options.relative_path_base = image_folder
+    validation_options.return_data = True
+    
+    results = validate_batch_results(output_file,validation_options)
+    assert len(results['validation_results']['errors']) == 0
+    assert len(results['validation_results']['warnings']) == 0
+    
+    
+    #%% Preview results
+    
+    from megadetector.postprocessing.postprocess_batch_results import \
+        PostProcessingOptions, process_batch_results
+        
+    postprocessing_options = PostProcessingOptions()
+    
+    postprocessing_options.md_results_file = output_file    
+    postprocessing_options.output_dir = r'g:\temp\serengeti-conversion-preview'
+    postprocessing_options.image_base_dir = image_folder
+    postprocessing_options.confidence_threshold = 0.2
+    postprocessing_options.num_images_to_sample = None
+    postprocessing_options.viz_target_width = 1280
+    postprocessing_options.line_thickness = 4    
+    postprocessing_options.parallelize_rendering_n_cores = 10
+    postprocessing_options.parallelize_rendering_with_threads = True
+        
+    postprocessing_results = process_batch_results(postprocessing_options)
+    
+    from megadetector.utils.path_utils import open_file
+    open_file(postprocessing_results.output_html_file)
+    
+    
+#%% Command-line driver
+
+def main():
+        
+    parser = argparse.ArgumentParser()
+    parser.add_argument('result_csv_file', type=str, 
+                        help='csv file containing AI results')
+    parser.add_argument('class_mapping_csv_file', type=str, 
+                        help='csv file containing class mappings (with columns CLASS, SPECIES)')
+    parser.add_argument('image_folder', type=str, 
+                        help='folder containing the images referred to in [result_csv_file]')
+    parser.add_argument('--output_file', type=str, default=None,
+                        help='.json file to which we should write results (defaults to [result_csv_file].json)')
+    
+    if len(sys.argv[1:])==0:
+        parser.print_help()
+        parser.exit()
+        
+    args = parser.parse_args()
+    raic_csv_to_md_results(result_csv_file=args.result_csv_file,
+                           class_mapping_csv_file=args.class_mapping_csv_file,
+                           image_folder=args.image_folder,
+                           output_file=args.output_file)
+
+if __name__ == '__main__':    
+    main()

megadetector/data_management/importers/zamba_results_to_md_results.py

@@ -14,8 +14,7 @@
  corrected_label
 
  Because the MD results file fundamentally stores detections, what we'll
- actually do is created bogus detections that fill the entire image.  Detection
- coordinates are not currently used in Timelapse video video anyway.
+ actually do is create bogus detections that fill the entire image.
 
  There is no special handling of empty/blank categories; because these results are
  based on a classifier, rather than a detector (where "blank" would be the absence of

megadetector/data_management/lila/create_lila_test_set.py

@@ -16,8 +16,6 @@ import random
 from megadetector.data_management.lila.lila_common import \
     read_lila_metadata, read_metadata_file_for_dataset
 
-from megadetector.utils.url_utils import download_url
-
 n_empty_images_per_dataset = 1
 n_non_empty_images_per_dataset = 1
 
@@ -48,6 +46,8 @@ for ds_name in metadata_table.keys():
 
 #%% Choose images from each dataset
 
+# Takes ~60 seconds
+
 # ds_name = (list(metadata_table.keys()))[0]
 for ds_name in metadata_table.keys():
 
@@ -102,10 +102,12 @@ for ds_name in metadata_table.keys():
 
 #%% Convert to URLs
 
+preferred_cloud = 'gcp'
+
 # ds_name = (list(metadata_table.keys()))[0]
 for ds_name in metadata_table.keys():
 
-    base_url = metadata_table[ds_name]['image_base_url']
+    base_url = metadata_table[ds_name]['image_base_url_' + preferred_cloud]
     assert not base_url.endswith('/')
     
     # Retrieve image file names
@@ -123,14 +125,14 @@ for ds_name in metadata_table.keys():
 # ...for each dataset
 
 
-#%% Download those image files
+#%% Download image files (prep)
+
+url_to_target_file = {}
 
-# TODO: trivially parallelizable
-#
 # ds_name = (list(metadata_table.keys()))[0]
 for ds_name in metadata_table.keys():
 
-    base_url = metadata_table[ds_name]['image_base_url']
+    base_url = metadata_table[ds_name]['image_base_url_' + preferred_cloud]
     assert not base_url.endswith('/')
     base_url += '/'
         
@@ -142,11 +144,23 @@ for ds_name in metadata_table.keys():
         assert base_url in url
         output_file_relative = ds_name.lower().replace(' ','_') + '_' + url.replace(base_url,'').replace('/','_').replace('\\','_')
         output_file_absolute = os.path.join(output_dir,output_file_relative)
-        try:
-            download_url(url, destination_filename=output_file_absolute, force_download=False, verbose=True)
-        except Exception as e:
-            print('\n*** Error downloading {} ***\n{}'.format(url,str(e)))
+        url_to_target_file[url] = output_file_absolute
         
     # ...for each url
     
 # ...for each dataset
+
+
+#%% Download image files (execution)
+
+from megadetector.utils.url_utils import parallel_download_urls
+
+download_results = parallel_download_urls(url_to_target_file,
+                                          verbose=False,
+                                          overwrite=False,
+                                          n_workers=20,
+                                          pool_type='thread')
+
+# r = download_results[0]
+for r in download_results:
+   assert r['status'] in ('skipped','success')

megadetector/data_management/lila/download_lila_subset.py

@@ -22,7 +22,8 @@ for s in lila_base_urls.values():
     assert s.endswith('/')
 
 # If any of these strings appear in the common name of a species, we'll download that image
-species_of_interest = ['grey fox','gray fox','cape fox','red fox','kit fox']
+# species_of_interest = ['grey fox','gray fox','cape fox','red fox','kit fox']
+species_of_interest = ['bear']
 
 # We'll write images, metadata downloads, and temporary files here
 lila_local_base = os.path.expanduser('~/lila')
@@ -45,7 +46,7 @@ random.seed(0)
 
 #%% Download and open the giant table of image URLs and labels
 
-# Takes ~60 seconds to download, unzip, and open
+# Takes ~2 minutes to download, unzip, and open
 df = read_lila_all_images_file(metadata_dir)
 
 
@@ -144,6 +145,12 @@ download_results = parallel_download_urls(url_to_target_file=url_to_target_file,
                                           pool_type='thread')
 
 
+#%% Open output folder
+
+from megadetector.utils.path_utils import open_file
+open_file(output_dir)
+
+
 #%% Scrap
 
 if False:

megadetector/data_management/lila/generate_lila_per_image_labels.py

@@ -57,6 +57,7 @@ ds_name_to_annotation_level['Channel IslandsCamera Traps'] = 'image'
 ds_name_to_annotation_level['WCS Camera Traps'] = 'sequence'
 ds_name_to_annotation_level['Wellington Camera Traps'] = 'sequence'
 ds_name_to_annotation_level['NACTI'] = 'unknown'
+ds_name_to_annotation_level['Seattle(ish) Camera Traps'] = 'image'
 
 known_unmapped_labels = set(['WCS Camera Traps:#ref!'])
 
@@ -103,7 +104,7 @@ for i_row,row in taxonomy_df.iterrows():
 
 #%% Process annotations for each dataset
 
-# Takes several hours
+# Takes a few hours
 
 # The order of these headers needs to match the order in which fields are added later in this cell;
 # don't mess with this order.
@@ -173,7 +174,7 @@ with open(output_file,'w',encoding='utf-8',newline='') as f:
             expected_annotation_level = None
                     
         # im = images[10]
-        for i_image,im in enumerate(images):
+        for i_image,im in tqdm(enumerate(images),total=len(images)):
             
             if (debug_max_images_per_dataset is not None) and (debug_max_images_per_dataset > 0) \
                 and (i_image >= debug_max_images_per_dataset):

megadetector/data_management/lila/test_lila_metadata_urls.py

@@ -52,6 +52,8 @@ print('Loaded metadata URLs for {} datasets'.format(len(metadata_table)))
 
 #%% Download and extract metadata and MD results for each dataset
 
+# Takes ~60 seconds if everything needs to beo downloaded and unzipped
+
 for ds_name in metadata_table.keys():    
 
     # Download the main metadata file for this dataset
@@ -73,10 +75,12 @@ for ds_name in metadata_table.keys():
                                                json_url=md_results_url,
                                                force_download=force_download)
 
+# ...for each dataset
+
 
 #%% Build up a list of URLs to test
 
-# Takes ~15 mins, since it has to open all the giant .json files
+# Takes ~15 mins, since it has to open all the giant .json files.
 
 url_to_source = {}
 

megadetector/data_management/read_exif.py

@@ -16,7 +16,7 @@ path.  No attempt is made to be consistent in format across the two approaches.
 import os
 import subprocess
 import json
-from datetime import date, datetime
+from datetime import datetime
 
 from multiprocessing.pool import ThreadPool as ThreadPool
 from multiprocessing.pool import Pool as Pool
@@ -27,6 +27,7 @@ from PIL import Image, ExifTags
 from megadetector.utils.path_utils import find_images, is_executable
 from megadetector.utils.ct_utils import args_to_object
 from megadetector.utils.ct_utils import image_file_to_camera_folder
+from megadetector.data_management.cct_json_utils import write_object_with_serialized_datetimes
 
 debug_max_images = None
 
@@ -88,8 +89,8 @@ class ReadExifOptions:
 
 class ExifResultsToCCTOptions:
     """
-    Options controlling the behavior of exif_results_to_cct() (which reformats the datetime information)
-    extracted by read_exif_from_folder().
+    Options controlling the behavior of exif_results_to_cct() (which reformats the datetime 
+    information) extracted by read_exif_from_folder().
     """
     
     def __init__(self):
@@ -730,9 +731,11 @@ def exif_results_to_cct(exif_results,cct_output_file=None,options=None):
     
     Args:
         exif_results (str or list): the filename (or loaded list) containing the results
-          from read_exif_from_folder
-        cct_file (str,optional): the filename to which we should write COCO-Camera-Traps-formatted
-          data
+            from read_exif_from_folder
+        cct_output_file (str,optional): the filename to which we should write 
+            COCO-Camera-Traps-formatted data
+        options (ExifResultsToCCTOptions, optional): options guiding the generation
+            of the CCT file, particularly location mapping
           
     Returns:
         dict: a COCO Camera Traps dict (with no annotations).
@@ -811,15 +814,8 @@ def exif_results_to_cct(exif_results,cct_output_file=None,options=None):
     d['annotations'] = []
     d['categories'] = []
     
-    def json_serialize_datetime(obj):
-        if isinstance(obj, (datetime, date)):
-            return obj.isoformat()
-        raise TypeError('Object {} (type {}) not serializable'.format(
-            str(obj),type(obj)))
-        
     if cct_output_file is not None:
-        with open(cct_output_file,'w') as f:
-            json.dump(d,f,indent=1,default=json_serialize_datetime)
+        write_object_with_serialized_datetimes(d,cct_output_file)
     
     return d
 

megadetector/data_management/rename_images.py

@@ -1,6 +1,6 @@
 """
 
-rename_images.py.py
+rename_images.py
 
 Copies images from a possibly-nested folder structure to a flat folder structure, including EXIF
 timestamps in each filename.  Loosely equivalent to camtrapR's imageRename() function.