initial commit

hash.py

@@ -0,0 +1,110 @@
+from __future__ import division
+import Image
+import os
+
+'''
+This shows how to use the data timelines to implement a "Find Similar Clips"
+feature, i.e. how to compute sequences and their hashes. There are two modes:
+similar shapes and similar colors.
+'''
+
+ZONE_INDEX = []
+for pixel_index in range(64):
+    x, y = pixel_index % 8, int(pixel_index / 8)
+    ZONE_INDEX.append(int(x / 2) + int(y / 4) * 4)
+
+def get_hash(image, mode, debug=False):
+    if mode == 'color':
+        # divide the image into 8 zones:
+        # 0 0 1 1 2 2 3 3
+        # 0 0 1 1 2 2 3 3
+        # 0 0 1 1 2 2 3 3
+        # 0 0 1 1 2 2 3 3
+        # 4 4 5 5 6 6 7 7
+        # 4 4 5 5 6 6 7 7
+        # 4 4 5 5 6 6 7 7
+        # 4 4 5 5 6 6 7 7
+        image_data = image.getdata()
+        image_hash = 0
+        zone_values = []
+        for zone_index in range(8):
+            zone_values.append([])
+        for pixel_index, pixel_value in enumerate(image_data):
+            zone_values[ZONE_INDEX[pixel_index]].append(pixel_value)
+        for zone_index, pixel_values in enumerate(zone_values):
+            # get the mean for each color channel
+            mean = map(lambda x: int(round(sum(x) / 8)), zip(*pixel_values))
+            # store the mean color of each zone as an 8-bit value:
+            # RRRGGGBB
+            color_index = sum((
+                int(mean[0] / 32) << 5,
+                int(mean[1] / 32) << 2,
+                int(mean[2] / 64)
+            ))
+            image_hash += color_index * pow(2, zone_index * 8)
+    elif mode == 'shape':
+        image_data = image.convert('L').getdata()
+        image_mean = sum(image_data) / 64
+        image_hash = 0
+        for pixel_index, pixel_value in enumerate(image_data):
+            if pixel_value > image_mean:
+                image_hash += pow(2, pixel_index)
+    return image_hash
+
+def get_sequences(path):
+    modes = ['color', 'shape']
+    sequences = {}
+    for mode in modes:
+        sequences[mode] = []
+    fps = 25
+    position = 0
+    file_names = filter(lambda x: 'timelinedata8p' in x, os.listdir(path))
+    file_names = sorted(file_names, key=lambda x: int(x[14:-4]))
+    file_names = map(lambda x: path + x, file_names)
+    for file_name in file_names:
+        timeline_image = Image.open(file_name)
+        timeline_width = timeline_image.size[0]
+        for x in range(0, timeline_width, 8):
+            frame_image = timeline_image.crop((x, 0, x + 8, 8))
+            for mode in modes:
+                frame_hash = get_hash(frame_image, mode)
+                if position == 0 or frame_hash != sequences[mode][-1]['hash']:
+                    if position > 0:
+                        sequences[mode][-1]['out'] = position
+                    sequences[mode].append({'in': position, 'hash': frame_hash})
+            position += 1 / fps
+    for mode in modes:
+        sequences[mode][-1]['out'] = position
+    return sequences
+
+if __name__ == '__main__':
+    from time import time
+    start = time()
+    sequences = get_sequences('../tiles/0084628/')
+    #sequences = get_sequences('../tiles/0097514/')
+    print 'get_sequences() took', time() - start , 'seconds'
+    for mode in ['color', 'shape']:
+        hashes = []
+        index = {}
+        for sequence in sequences[mode]:
+            if sequence['hash'] > 0:
+                if not sequence['hash'] in index:
+                    index[sequence['hash']] = len(hashes)
+                    hashes.append({
+                        'hash': sequence['hash'],
+                        'sequences': []
+                    })
+                hashes[index[sequence['hash']]]['sequences'].append({
+                    'in': sequence['in'],
+                    'out': sequence['out']
+                })
+        hashes = filter(lambda x: len(x['sequences']) > 1, hashes)
+        print '-' * 64
+        print mode
+        print '-' * 64
+        #for h in sorted(hashes, key=lambda x: -len(x['sequences'])):
+        for h in sorted(hashes, key=lambda x: x['sequences'][-1]['out'] - x['sequences'][0]['in']):
+            print h['hash'], len(h['sequences']), ', '.join(
+                map(lambda x: '%.2f-%.2f' % (x['in'], x['out']), h['sequences'])
+            )
+    

oxtimelinesutils.py

@@ -0,0 +1,258 @@
+from __future__ import division
+import Image
+import json
+import math
+import os
+import re
+
+'''
+This is an implementation of a join_tiles function for new-style timelines.
+Timelines of files will be read from source_paths, the timeline of the item will
+be written to target_path.
+'''
+
+def join_tiles(source_paths, target_path):
+
+    def divide(num, by):
+        # divide(100, 3) -> [33, 33, 34]
+        arr = []
+        div = int(num / by)
+        mod = num % by
+        for i in range(int(by)):
+            arr.append(div + (i > by - 1 - mod))
+        return arr
+
+    def get_file_info(file_name):
+        for mode in modes:
+            if re.match('^timeline' + mode + '64p\d+\.jpg', file_name):
+                return {
+                    'file': file_name,
+                    'mode': mode,
+                    'index': int(file_name[11 + len(mode):-4])
+                }
+        return None
+
+    def save_and_open(data):
+        # whenever a large tile is done or needed,
+        # this function saves the previous large tile
+        # (if any) and opens the next one (if any).
+        # in between, whenever required, small tiles
+        # are opened, rendered and saved, and the
+        # large full tile is being generated.
+        # 'keyframes' are only rendered in large size,
+        # 'keyframeswide' only resized to small size.
+        image_mode = 'L' if mode == 'audio' else 'RGB'
+        small_mode = 'keyframes' if mode == 'keyframeswide' else mode
+        large_tile_i = int(target_w / large_tile_w)
+        # save previous large tile
+        if large_tile_i > 0:
+            large_tile_i -= 1
+            if mode != 'keyframeswide':
+                image_file = '%stimeline%s%dp%d.jpg' % (
+                    target_path, mode, large_tile_h, large_tile_i
+                )
+                data['target_images']['large'].save(image_file)
+                print image_file
+            if mode != 'keyframes':
+                # open small tile
+                small_tile_i = int(large_tile_i / 60)
+                small_tile_x = (large_tile_i % 60) * 60
+                if small_tile_x == 0:
+                    if small_tile_i < small_tile_n - 1:
+                        w = small_tile_w
+                    else:
+                        w = small_tile_last_w
+                    data['target_images']['small'] = Image.new(image_mode, (w, small_tile_h))
+                # paste large tile into small tile
+                w = 60 if large_tile_i < large_tile_n - 1 else small_tile_last_w % 60
+                data['target_images']['large'] = data['target_images']['large'].resize(
+                    (w, small_tile_h), Image.ANTIALIAS
+                )
+                data['target_images']['small'].paste(
+                    data['target_images']['large'], (small_tile_x, 0)
+                )
+                # save small tile
+                if small_tile_x == small_tile_w - 60 or large_tile_i == large_tile_n - 1:
+                    image_file = '%stimeline%s%dp%d.jpg' % (
+                        target_path, small_mode, small_tile_h, small_tile_i
+                    )
+                    data['target_images']['small'].save(image_file)
+                    print image_file
+            if mode == 'antialias':
+                # render full tile
+                resized = data['target_images']['large'].resize((
+                    data['full_tile_widths'][0], large_tile_h
+                ), Image.ANTIALIAS)
+                data['target_images']['full'].paste(resized, (data['full_tile_offset'], 0))
+                data['full_tile_offset'] += data['full_tile_widths'][0]
+                data['full_tile_widths'] = data['full_tile_widths'][1:]
+            large_tile_i += 1
+        # open next large tile
+        if large_tile_i < large_tile_n:
+            w = large_tile_w if large_tile_i < large_tile_n - 1 else large_tile_last_w
+            data['target_images']['large'] = Image.new(image_mode, (w, large_tile_h))
+
+    data = {}
+    fps = 25
+    large_tile_w, large_tile_h = 1500, 64
+    small_tile_w, small_tile_h = 3600, 16
+    full_tile_w = 1920
+    modes = ['antialias', 'slitscan', 'keyframes', 'keyframeswide', 'audio']
+    source_files = {}
+    for mode in modes:
+        source_files[mode] = []
+
+    # read files
+    durations = [0] * len(source_paths)
+    frame_n = 0
+    for i, path in enumerate(source_paths):
+        file_info = map(get_file_info, os.listdir(path))
+        file_info = filter(lambda x: x != None, file_info)
+        for info in sorted(file_info, key=lambda x: x['index']):
+            mode = info['mode']
+            source_files[mode].append(path + info['file'])
+            if mode == modes[0]:
+                width = Image.open(source_files[mode][-1]).size[0]
+                durations[i] += width / fps
+                frame_n += width
+    large_tile_n = int(math.ceil(frame_n / large_tile_w)) 
+    large_tile_last_w = frame_n % large_tile_w
+    small_tile_n = int(math.ceil(frame_n / fps / small_tile_w)) 
+    small_tile_last_w = int(math.ceil(frame_n / fps)) % small_tile_w
+
+    # open full timeline
+    if large_tile_n == 1:
+        data['full_tile_widths'] = [large_tile_last_w]
+    else:
+        w = full_tile_w
+        n = large_tile_n
+        if large_tile_last_w < large_tile_w:
+            factor = full_tile_w / frame_n
+            last_w = int(round(large_tile_last_w * factor))
+            w -= last_w
+            n -= 1
+        data['full_tile_widths'] = divide(w, n)
+        if large_tile_last_w < large_tile_w:
+            data['full_tile_widths'].append(last_w)
+    data['full_tile_offset'] = 0
+    full_tile_image = Image.new('RGB', (full_tile_w, large_tile_h))
+
+    # main loop
+    data['target_images'] = {'large': None, 'small': None, 'full': full_tile_image}
+    for mode in modes:
+        target_w = 0
+        for source_file in source_files[mode]:
+            source_image = Image.open(source_file)
+            source_w = source_image.size[0]
+            target_x = target_w % large_tile_w
+            if target_x == 0:
+                save_and_open(data)
+            data['target_images']['large'].paste(source_image, (target_x, 0))
+            target_w += source_w
+            if target_x + source_w > large_tile_w:
+                # target tile overflows into next source tile
+                save_and_open(data)
+                target_x -= large_tile_w
+                data['target_images']['large'].paste(source_image, (target_x, 0))
+        target_w += source_w
+        save_and_open(data)
+
+    # save full timelines
+    image_file = '%stimelineantialias%dp.jpg' % (target_path, large_tile_h)
+    data['target_images']['full'].save(image_file)
+    print image_file
+    image_file = '%stimelineantialias%dp.jpg' % (target_path, small_tile_h)
+    data['target_images']['full'].resize(
+        (full_tile_w, small_tile_h), Image.ANTIALIAS
+    ).save(image_file)
+    print image_file
+
+    # join cuts
+    cuts = []
+    offset = 0
+    for i, path in enumerate(source_paths):
+        f = open(path + 'cuts.json', 'r')
+        path_cuts = json.load(f)
+        f.close()
+        if i > 0:
+            cuts.append(offset)
+        for cut in path_cuts:
+            cuts.append(offset + cut)
+        offset += durations[i]
+    f = open(target_path + 'cuts.json', 'w')
+    # avoid float rounding artefacts
+    f.write('[' + ', '.join(map(lambda x: '%.2f' % x, cuts)) + ']')
+    f.close()    
+
+
+def split_tiles(path, paths, durations):
+
+    def is_timeline_file(file_name):
+        return file_name.startswith('timeline') and file_name.endswith('.png')
+
+    file_names = filter(is_timeline_file, os.listdir(path))
+    tiles = {}
+    for file_name in file_names:
+        mode = re.split('\d+', file_name[8:])[0]
+        print file_name, mode
+        split = re.split('[a-z]+', file_name[8 + len(mode):-4])
+        height, index = map(lambda x: int(x) if len(x) else -1, split)
+        if not mode in tiles:
+            tiles[mode] = {}
+        if not height in tiles[mode]:
+            tiles[mode][height] = 0
+        if index + 1 > tiles[mode][height]:
+            tiles[mode][height] = index + 1
+    print tiles
+
+    # for each mode
+    for mode in tiles:
+        image_mode = 'L' if mode == 'audio' else 'RGB'
+        # and for each size of that mode
+        for i, height in enumerate(tiles[mode]):
+            tile_width = 1500 if i == 0 else 3600
+            px_per_sec = 25 if i == 0 else 1
+            target_images = []
+            target_data = []
+            # and for each split item
+            for item_index, duration in enumerate(durations):
+                tile_index = 0
+                px = int(math.ceil(duration * px_per_sec))
+                # create a flat list of all target images   
+                # (and store the split item and tile index)
+                while px:
+                    width = tile_width if px > tile_width else px
+                    target_images.append(
+                        Image.new(image_mode, (width, height))
+                    )
+                    target_data.append(
+                        {'item': item_index, 'tile': tile_index}
+                    )
+                    tile_index += 1
+                    px -= width
+            target_index = 0
+            offset = 0
+            # for each source tile
+            for source_index in range(tiles[mode][height]):
+                source_image = Image.open('%stimeline%s%dp%d.png' % (
+                    path, mode, height, source_index
+                ))
+                source_width = source_image.size[0]
+                target_width = target_images[target_index].size[0]
+                target_images[target_index].paste(source_image, (offset, 0))
+                # paste it into as many target tiles as needed
+                while source_width + offset > target_width:
+                    offset -= target_width
+                    target_index += 1
+                    target_width = target_images[target_index].size[0]
+                    target_images[target_index].paste(source_image, (offset, 0))
+            for i, target_image in enumerate(target_images):
+                file_name = '%stimeline%s%dp%d' % (
+                    paths[target_data[i]['item']], mode, height, target_data[i]['tile']
+                )
+                # target_image.save(file_name)
+                print file_name, target_image.size
+
+if __name__ == '__main__':
+    join_tiles(['../tiles/0196499/1/', '../tiles/0196499/2/', '../tiles/0196499/3/'], '../tiles/0196499/')
+    # split_tiles('../tiles/0097514/', ['a/', 'b/', 'c/', 'd/', '/e'], [1000, 666.64, 666.68, 666.68, 29.96])