def compute_sim(self, key):
print 'compute similarities for', self.img
self.message('Compute similarities:\n\n{}\nx{} imgs'.format(
self.img,
len(picture.pictures())))
sims = {}
for p in picture.pictures():
if p != self.img:
sim = self.img.similarity(p)
sims[p] = sim
minsim = min(sims.values())
maxsim = max(sims.values())
thresh = maxsim - (maxsim-minsim)/3.
res = []
for p,s in sims.items():
if s > thresh:
picture.connect(self.img, p, s)
res.append(p)
#x=0
#for q in res[:10]:
#img = self.load_thmb(q)
#self.cnv.create_image((x,780),
#anchor=tk.SW, image=img)
#self.cur_imgs.append(img)
#x += img.width()
print 'done. found {} images.'.format(len(res))
self.merge_candidates = self.merge_cand()
self.redraw=True
return res
def cluster_mode(self, key):
if self.mode in [Browser.BROWSE, Browser.BLOG, Browser.POPULAR]:
#if len(self.clusters)<1:
self.message('Clustering...')
self.clusters = index.clustering(picture.pictures(),
len(picture.pictures())-100)
self.mode='cluster'
for c in self.clusters:
if self.img in c:
self.pool = c
self.redraw=True
else:
self.mode = Browser.BROWSE
self.repool()
self.redraw=True
def simpairs():
res=[]
imgs=picture.pictures()
print 'Begin computing similarities between {} images'.format(
len(imgs))
# BEGIN
for i in range(len(imgs)):
p=imgs[i]
for j in range(i+1,len(imgs)):
q=imgs[j]
sim=p.similarity(q)
if sim>.5:
res.append((p,q,sim))
picture.connect(p,q,sim)
#if i%100==0:
#print '\t{}\t/\t{}'.format(i, len(imgs)),
print '\tDone!\t\t\t'
f=open('html/.twins.html','w')
f.write('<html>\n<body>')
res.sort(key=lambda t:t[2], reverse=True)
for p,q,sim in res[:500]:
if sim > .9:
f.write('<h4>{} and {}: {}</h4>\n'.format(p.name,q.name,sim))
f.write('<b>{} versus {}: </b><br/>\n'.format(p.info,q.info,))
if len(p.sources)>0 and len(q.sources)>0:
f.write('<i>{} and {}: </i><br/>\n'.format(p.origin.name,q.origin.name,))
f.write('<img src="../{}"/><img src="../{}"/><br/>\n'.format(
p.location, q.location))
f.write('</body>\n</html>\n')
f.close()
print 'Done'
return res
def __init__(self, root):
self.cur_imgs = [] # backup references against garbage coll.
self.img_reg = {} # registry to avoid disk access
self.thmb_reg = {} # thumbnail registry to avoid resize
self.preview_reg = {} # preview registry to avoid resize
self.mode = Browser.BROWSE
self.redraw = False # set true to redraw gui
self.changes = False # changes to be saved?
self.new_votes = set() # keep track of new ratings
self.pool=[]
# TODO: selection: GUI indicator must be present for selection state,
# selection must be viewable in its entirety in dedicated view mode,
# selection should probably be of type set.
self.selection=[] # selection for exports or queries
# compare new pictures with favorites
favs = picture.starred()[:10]
newsies = [p for p in picture.pictures() if p.reviewed < 1]
if len(newsies)>0 and len(favs)>0:
print 'calculating similarities between {} new images and highest rated'.format(
len(newsies))
self.new_votes = set(favs)
for n in newsies:
for p in favs:
if n != p:
sim = p.similarity(n)
if sim > .6:
picture.connect(n, p, sim)
print 'done doing that.'
# merge candidates for currentlt selected image
self.merge_candidates=[]
# img clusters
self.clusters=[]
# init img tracking
self.repool()
#pics = sorted(pics, key=lambda p:p.rating)
# repopulate history
self.hist = []
for p in picture.last_reviewed()[:50]:
if util.days_since(p.reviewed)<1.5:
self.hist.append(p)
if p in self.pool:
self.pool.remove(p)
# choose image to displ
if len(self.hist)<1 and len(self.pool)>0:
self.choose(self.pool.pop()) # current image as Pict object
self.hist = [self.img] # history of recent images
elif len(self.hist)>0:
self.img = self.hist[0]
# self.trash
# key: pict instance, value: (path, image)
self.trash = {}
# canvas
self.cnv = tk.Canvas(root, bg="black")
self.cnv.pack(side='top', fill='both', expand='yes')
# put a button on the image panel to test it
#self.button = tk.Button(self.cnv, text='button2')
#self.button.pack(side='top')
self.merge_candidates = self.merge_cand()
self.display()
def import_dump(self, key):
if not os.path.exists('exports'):
os.mkdir('exports')
imgs=[]
srcimgs=[]
blgs=[]
#TODO: find all xml files in folder
if os.path.exists('exports/images.xml'):
#TODO: download image from original location
imgrec = util.inout.loadImages('exports/images.xml')
srcimgs = [picture.opendump(rec) for rec in imgrec]
# FIXME: workaround wegen offline: bilder werden komplett mitkopiert
for p in srcimgs:
if not os.path.exists('images/{}'.format(p.filename)):
os.rename('exports/{}'.format(p.filename), 'images/{}'.format(p.filename))
imgs.append(p)
# remove images xml records so that missing images cannot be imported again
os.remove('exports/images.xml')
if os.path.exists('exports/blogs.xml'):
blgrec = util.inout.loadBlogs('exports/blogs.xml')
blgs = [tumblr.opendump(rec) for rec in blgrec]
# reify image references made by blogs
for t in blgs:
index.clean_img_refs(t)
# remove xml file because import is successful
os.remove('exports/blogs.xml')
# now that we have our blogs imported, we can reify blog/img references
# in img instances
# TODO: reification of source blogs, interblog references, interimg links!!
for p in srcimgs:
index.clean_sources(p)
p.clean_links()
self.message('imported {} images and {} blogs.'.format(
len(srcimgs), len(blgs)))
# compute similarities with present images
self.message('compute similarities with present images..')
bestsim=0 # sim stat
for p in imgs:
sims = {}
for q in picture.pictures():
if q != p:
sims[q] = p.similarity(q)
minsim,maxsim = (min(sims.values()), max(sims.values()))
p.relates.update({q:s for q,s in sims.items()
if s > maxsim-(maxsim-minsim)/3})
# keep track of best match
bestsim = max(bestsim, maxsim)
# remove xml files (now left without their actual images...)
# and repool
if len(imgs)>0:
self.pool = imgs
self.redraw = True
self.message('\n'.join(['Imported {} image records with {} new images'.format(
len(srcimgs), len(imgs)),
'featured by {} blogs.'.format(len(blgs)),
'Highest similarity between old and new image was {:.2f}'.format(bestsim)]),
confirm=True)
#!/usr/bin/python import index import weave.picture as picture import weave.tumblr as tumblr import util.cluster as clustering index.load() index.simpairs() index.save() pics = picture.pictures() pics = sorted(pics, key=lambda p: len(p.relates)) index.stumblr(pics[-1], "walk.html") clusters = clustering.linkage(pics, len(pics) / 10, mode=clustering.LINK_NORM) groups = [c.members for c in clusters] index.savegroups(groups, "cluster.html")
def pictures(): return picture.pictures()