"""Offline geometric candidates, palette fitting and perceptual quality checks.""" import io, zlib import numpy as np from scipy.ndimage import uniform_filter, distance_transform_edt from scipy.spatial import cKDTree from PIL import Image def initialize(records, originals): global R, O, NE, N, M, RGBA, IDX, ALPHA R = records; O = originals; NE = len(O); N = len(R) M = {d['id']: NE + i for i, d in enumerate(R[NE:])} RGBA = np.array([d['rgba'] for d in R[:NE]]) for i, d in enumerate(R[:NE]): if d['mask'] != 65535: RGBA[i, :, :, 3] = R[M[d['mask']]]['rgba'][:, :, 0] IDX = np.array([d['idx'] for d in R]); ALPHA = RGBA[:, :, :, 3] def trans(a,t): if t&4:a=a.swapaxes(0,1) if t&2:a=a[::-1] if t&1:a=a[:,::-1] return a def features(a): alpha=a[:,:,3:]/255.;rgb=a[:,:,:3]*alpha g=rgb.mean(axis=2);g=g/(g.std()+10) return np.concatenate([g.reshape(16,4,16,4).mean(axis=(1,3)).ravel(),alpha.reshape(16,4,16,4).mean(axis=(1,3)).ravel()*2]).astype(np.float32) def make_candidates(): global C F=np.array([features(a) for a in RGBA]);TREE=cKDTree(F) C=[[] for _ in range(NE)] for t in [0,1,2,3,4,5,6,7]: f=np.array([features(trans(a,t)) for a in RGBA]);_,js=TREE.query(f,k=9) for j,row in enumerate(js): for i in row: if i!=j:C[i].append((j,t)) # Add direct near candidates; mirror query inversion above is exact involution only for 0..4,7. # Here the candidate is transformed j, queried against normal i, so all eight are correct. return C def metrics(a,o): a=a.astype(np.float32)/255;o=o.astype(np.float32)/255 ms=ss=p99=0. for bg in [0,1]: x=a[:,:,:3]*a[:,:,3:]+bg*(1-a[:,:,3:]);y=o[:,:,:3]*o[:,:,3:]+bg*(1-o[:,:,3:]);err=(x-y)**2 ms+=float(np.mean(err));p99+=float(np.quantile(err,.99)) mx=uniform_filter(x,size=(7,7,1));my=uniform_filter(y,size=(7,7,1));vx=uniform_filter(x*x,size=(7,7,1))-mx*mx;vy=uniform_filter(y*y,size=(7,7,1))-my*my;cv=uniform_filter(x*y,size=(7,7,1))-mx*my ss+=float(np.mean(((2*mx*my+.0001)*(2*cv+.0009))/((mx*mx+my*my+.0001)*(vx+vy+.0009)))) return ms/2,ss/2,p99/2 def fit(i,j,t): ix=trans(IDX[j],t);n=len(R[j]['pal']);a=ALPHA[i].astype(np.float32)/255;o=O[i].astype(np.float32);ao=o[:,:,3]/255 weight=a*a;wanted=np.zeros((64,64,3),np.float32);valid=a>0 wanted[valid]=(o[valid,:3]*ao[valid,None]+127.5*(a[valid,None]-ao[valid,None]))/a[valid,None] den=np.bincount(ix.ravel(),weight.ravel(),minlength=n) pal=np.array([np.bincount(ix.ravel(),(wanted[:,:,c]*weight).ravel(),minlength=n)/np.maximum(den,1e-6) for c in range(3)]).T pal=np.clip(np.round(pal),0,255).astype(np.uint8) output=np.empty((64,64,4),np.uint8);output[:,:,:3]=pal[ix];output[:,:,3]=ALPHA[i] return output,pal def template_job(i): if R[i]['mask']==65535:return [],None ref=metrics(RGBA[i],O[i]);options=[] for j,t in C[i]: if len(R[j]['pal'])>150:continue a,pal=fit(i,j,t);met=metrics(a,O[i]);good=(met[0]<=ref[0] and met[1]>=ref[1] and met[2]<=ref[2]) if good:options.append((len(pal)*3,j,t,pal,met)) options.sort(key=lambda x:x[0]);return options[:8],ref def half_features(a): alpha=a[:,:,3:]/255.;g=(a[:,:,:3]*alpha).mean(axis=2);g/=g.std()+10 return np.concatenate([g.reshape(16,4,8,4).mean(axis=(1,3)).ravel(),alpha.reshape(16,4,8,4).mean(axis=(1,3)).ravel()*2]).astype(np.float32) def make_half_candidates(): global HC HC=[[[] for _ in range(NE)] for _ in range(2)] for side in [0,1]: target=np.array([half_features(a[:,side*32:(side+1)*32]) for a in RGBA]);tree=cKDTree(target) for t in [0,1]: f=np.array([half_features(trans(a,t)[:,side*32:(side+1)*32]) for a in RGBA]);_,js=tree.query(f,k=12) for j,row in enumerate(js): for i in row: if i!=j:HC[side][i].append((j,t)) return HC def fit_half(i,j,t,side): x=side*32;ix=trans(IDX[j],t)[:,x:x+32];n=len(R[j]['pal']);a=ALPHA[i,:,x:x+32].astype(np.float32)/255;o=O[i,:,x:x+32].astype(np.float32);ao=o[:,:,3]/255 wt=a*a;valid=a>0;wanted=np.zeros((64,32,3),np.float32) wanted[valid]=(o[valid,:3]*ao[valid,None]+127.5*(a[valid,None]-ao[valid,None]))/a[valid,None] den=np.bincount(ix.ravel(),wt.ravel(),minlength=n) pal=np.array([np.bincount(ix.ravel(),(wanted[:,:,c]*wt).ravel(),minlength=n)/np.maximum(den,1e-6) for c in range(3)]).T pal=np.clip(np.round(pal),0,255).astype(np.uint8) out=np.empty((64,32,4),np.uint8);out[:,:,:3]=pal[ix];out[:,:,3]=ALPHA[i,:,x:x+32] return out,pal def half_template_job(i): if R[i]['mask']==65535:return [] halves=[] for side in [0,1]: x=side*32;original=O[i,:,x:x+32];ref=metrics(RGBA[i,:,x:x+32],original);opts=[] for j,t in HC[side][i]: if len(R[j]['pal'])>150:continue out,pal=fit_half(i,j,t,side);met=metrics(out,original) if met[0]<=ref[0] and met[1]>=ref[1] and met[2]<=ref[2]:opts.append((len(np.unique(pal,axis=0)),j,t,pal,met)) opts.sort(key=lambda x:x[0]);halves.append(opts[:4]) if not halves[0] or not halves[1]:return [] options=[] for l in halves[0]: for r in halves[1]: pal,inv=np.unique(np.concatenate([l[3],r[3]]),axis=0,return_inverse=True) if len(pal)>256:continue raw=inv.astype(np.uint8).tobytes()+bytes(4096);zz=zlib.compress(raw,9) options.append((len(pal)*3+len(zz),l[1],r[1],l[2],r[2],pal,zz,l[4],r[4])) return sorted(options,key=lambda x:x[0])[:6] def webp(a): b=io.BytesIO();Image.fromarray(a).save(b,format='WEBP',lossless=True,quality=100,method=6,exact=True);return b.getvalue() def mirror_job(i): a=IDX[i];visible=ALPHA[i]>0 if i=rr[1] and sc[2]<=rr[2] out.append((q,len(b),sc,good)) if good: chosen=(q,b,sc);break return chosen,rr,out def error(a,o): a=a.astype(np.float32)/255;o=o.astype(np.float32)/255 da=a[:,:,3:]-o[:,:,3:];rgb=a[:,:,:3]*a[:,:,3:]-o[:,:,:3]*o[:,:,3:] return ((rgb*rgb+(rgb-da)*(rgb-da))/2).mean(axis=2) def structure(a,o): a=a.astype(np.float32)/255;o=o.astype(np.float32)/255;s=0 for bg in [0,1]: x=a[:,:,:3]*a[:,:,3:]+bg*(1-a[:,:,3:]);y=o[:,:,:3]*o[:,:,3:]+bg*(1-o[:,:,3:]) mx=uniform_filter(x,size=(5,5,1));my=uniform_filter(y,size=(5,5,1));vx=uniform_filter(x*x,size=(5,5,1))-mx*mx;vy=uniform_filter(y*y,size=(5,5,1))-my*my;cv=uniform_filter(x*y,size=(5,5,1))-mx*my s+=(((2*mx*my+.0001)*(2*cv+.0009))/((mx*mx+my*my+.0001)*(vx+vy+.0009))).mean(axis=2) return s/2 def passes_global(image, original, reference_metrics): m = metrics(image, original) return m[0] <= reference_metrics[0] and m[1] >= reference_metrics[1] and m[2] <= reference_metrics[2] def passes_local(image, original, reference_error, reference_structure): e = uniform_filter(error(image, original), size=5) s = structure(image, original) return bool(np.all(e <= np.maximum(reference_error * 3, (8 / 255) ** 2)) and np.all(s >= reference_structure - 0.1))