def run_generate(args):
os.system(f'mkdir {outpath}/gen_{args.exp_name} > /dev/null 2>&1')
decoder = torch.load(
f"{outpath}/{args.model_name}/models/decoder_{args.load_epoch}.pt").to(
device)
encoder = torch.load(
f"{outpath}/{args.model_name}/models/encoder_{args.load_epoch}.pt").to(
device)
random.seed(args.rd_seed)
np.random.seed(args.rd_seed)
torch.manual_seed(args.rd_seed)
with torch.no_grad():
if args.mode == "eval_gen":
i = 0
miss = 0
while (i < args.num_gen):
print(f"Gen {i}")
try:
h0 = torch.randn(1, 1, args.hidden_dim).to(device)
prog, _ = run_eval_decoder(h0, decoder, True)
verts, faces = hier_execute(prog)
utils.writeObj(
verts, faces,
f"{outpath}/gen_{args.exp_name}/gen_{i}.obj")
utils.writeHierProg(
prog,
f"{outpath}/gen_{args.exp_name}/gen_prog_{i}.txt")
i += 1
except Exception as e:
print(f"Failed to generate prog with {e}")
miss += 1
print(f"Gen reject %: {miss / (args.num_gen + miss)}")
if args.mode == "eval_recon":
ind_file = f'data_splits/{args.category}/val.txt'
inds = getInds(ind_file)
for ind in tqdm(inds):
gtprog = utils.loadHPFromFile(f'{args.dataset_path}/{ind}.txt')
gtverts, gtfaces = hier_execute(gtprog)
shape = progToData(gtprog)
enc, _ = get_encoding(shape, encoder, mle=True)
prog, _ = run_eval_decoder(enc, decoder, False)
verts, faces = hier_execute(prog)
utils.writeObj(
verts, faces,
f"{outpath}/gen_{args.exp_name}/{ind}_recon.obj")
utils.writeObj(gtverts, gtfaces,
f"{outpath}/gen_{args.exp_name}/{ind}_gt.obj")
def prog_to_pc(prog):
verts, faces = hier_execute(prog)
for i in range(3):
verts[:, i] = verts[:, i] - verts[:, i].mean()
pc = utils.sample_surface(faces,
verts.unsqueeze(0),
num_samps,
return_normals=False)[0]
return pc
def calc_tab3():
ddir = sys.argv[1]
inds = os.listdir(ddir)
outs = []
if len(sys.argv) > 2:
inds = inds[:int(sys.argv[2])]
for ind in tqdm(inds):
if '.txt' in ind:
hp = utils.loadHPFromFile(f'{ddir}/{ind}')
verts, faces = hier_execute(hp)
else:
verts, faces = utils.loadObj(f'{ddir}/{ind}')
verts = torch.tensor(verts)
faces = torch.tensor(faces)
outs.append((verts, faces))
misses = 0.
results = {
'num_parts': [],
'rootedness': [],
'stability': [],
}
samples = []
for (verts, faces) in tqdm(outs):
results['num_parts'].append(verts.shape[0] / 8.0)
samples.append((verts, faces))
if check_rooted(verts, faces):
results['rootedness'].append(1.)
else:
results['rootedness'].append(0.)
if check_stability(verts, faces):
results['stability'].append(1.)
else:
results['stability'].append(0.)
for key in results:
if len(results[key]) > 0:
res = torch.tensor(results[key]).mean().item()
else:
res = 0.
results[key] = res
results['variance'] = eval_get_var(samples)
for key in results:
print(f"Result {key} : {results[key]}")
def main(ind):
sa = ShapeAssembly()
root_lines, has_mid = make_skel()
prog_lines = ['Assembly Program_0 {']
for b in root_lines:
prog_lines.append('\t'+b)
prog_lines.append('}')
RP = Program()
for l in root_lines:
RP.execute(l)
base_par = 'mid' if has_mid else 'top'
base_lines = make_base_prog(
RP.cuboids['base'],
RP.cuboids[base_par]
)
for i in range(len(prog_lines)):
prog_lines[i] = prog_lines[i].replace('base', 'Program_1')
prog_lines += base_lines
#for b in base_lines:
# prog_lines.append('\t'+b)
#prog_lines.append('}')
cube_count = -1
switches = []
for line in prog_lines:
if 'Cuboid' in line:
if not ('Program_' in line or "bbox" in line):
switches.append((
f'cube{cube_count}', line.split()[0]
))
if "bbox" in line:
cube_count = -1
cube_count += 1
for a, b in switches:
prog_lines = [line.replace(b,a) for line in prog_lines]
hier_prog = make_hier_prog(prog_lines)
verts, faces = hier_execute(hier_prog)
if check_rooted(verts, faces) and check_stability(verts, faces):
# writeObj(verts, faces, f'out_{ind}.obj')
writeHierProg(hier_prog, f"random_hier_data/{ind}.txt")
return True
return False
def prog_to_pc(prog, ns):
if prog['prog'] == []:
return torch.zeros((ns, 3))
verts, faces = hier_execute(prog)
for i in range(3):
verts[:, i] = verts[:, i] - verts[:, i].mean()
pc = utils.sample_surface(faces,
verts.unsqueeze(0),
ns,
return_normals=False)[0]
return pc
def eval_recon(outdir, data_inds):
decoder = torch.load(shapeAssembly_decoder).to(device)
decoder.eval()
encoder = PCEncoder()
encoder.load_state_dict(torch.load(point_cloud_encoder))
encoder.eval()
encoder.to(device)
os.system(f'mkdir {outdir}')
count = 0.
tdist = 0.
for ind in tqdm(data_inds):
pc_samp = torch.load(f'{point_cloud_folder}/{ind}.pts').to(device)
enc = encoder(pc_samp.unsqueeze(0))
prog, _ = run_eval_decoder(enc.unsqueeze(0), decoder, False)
verts, faces = hier_execute(prog)
utils.writeObj(verts, faces, f"{outdir}/{ind}.obj")
utils.writeHierProg(prog, f"{outdir}/{ind}.txt")
verts = verts.to(device)
faces = faces.to(device)
pred_samp = utils.sample_surface(faces, verts.unsqueeze(0), 10000,
True)
# Center PC
offset = (pc_samp.max(dim=0).values + pc_samp.min(dim=0).values) / 2
pc_samp -= offset
#utils.writeSPC(pc_samp,f'tar_pc_{ind}.obj')
#utils.writeSPC(pred_samp[0,:,:3],f'scripts/output/pred_pc_{ind}.obj')
pc_samp = pc_samp.repeat(1, 2).unsqueeze(0)
tdist += fscore.score(pred_samp.squeeze().T.unsqueeze(0),
pc_samp.squeeze().T.unsqueeze(0))
count += 1
print(f"Average F-score: {tdist/count}")
def main(dataset_path, outdir):
indices, progs = load_progs(dataset_path)
os.system(f'mkdir {outdir}')
os.system(f'mkdir {outdir}/valid')
os.system(f'mkdir {outdir}/non_valid')
count = 0
for ind, prog in tqdm(list(zip(indices, progs))):
count += 1
if count < 0:
continue
lc = simplifyHP(prog)
verts, faces = hier_execute(prog)
bbdims = torch.tensor([
float(a)
for a in re.split(r'[()]', prog['prog'][0])[1].split(',')[:3]
])
bb_viol = (verts.abs().max(dim=0).values / (bbdims / 2)).max()
try:
rooted = check_rooted(verts, faces)
except Exception as e:
print(f"Failed rooted check with {e}")
rooted = False
if DO_STABLE:
stable = check_stability(verts, faces)
else:
stable = True
if lc <= MAX_LEAVES and lc >= MIN_LEAVES and bb_viol < BB_THRESH and rooted and stable:
utils.writeHierProg(prog, f'{outdir}/valid/{ind}.txt')
utils.writeObj(verts, faces, f'{outdir}/valid/{ind}.obj')
else:
utils.writeHierProg(prog, f'{outdir}/non_valid/{ind}.txt')
utils.writeObj(verts, faces, f'{outdir}/non_valid/{ind}.obj')
def gen_metrics(gen_progs,
outpath,
exp_name,
epoch,
VERBOSE,
write_progs=True):
misses = 0.
results = {
'num_parts': [],
'rootedness': [],
'stability': [],
}
samples = []
for i, prog in enumerate(gen_progs):
try:
verts, faces = hier_execute(prog)
assert not torch.isnan(verts).any(), 'saw nan vert'
if write_progs:
utils.writeObj(
verts, faces,
f"{outpath}/{exp_name}/objs/gen/{epoch}_{i}.obj")
utils.writeHierProg(
prog, f"{outpath}/{exp_name}/programs/gen/{epoch}_{i}.txt")
results['num_parts'].append(verts.shape[0] / 8.0)
samples.append((verts, faces))
except Exception as e:
misses += 1.
if VERBOSE:
print(f"failed gen metrics for {i} with {e}")
continue
try:
if check_rooted(verts, faces):
results['rootedness'].append(1.)
else:
results['rootedness'].append(0.)
if check_stability(verts, faces):
results['stability'].append(1.)
else:
results['stability'].append(0.)
except Exception as e:
if VERBOSE:
print(f"failed rooted/stable with {e}")
for key in results:
if len(results[key]) > 0:
res = torch.tensor(results[key]).mean().item()
else:
res = 0.
results[key] = res
try:
results['variance'] = eval_get_var(samples)
except Exception as e:
results['variance'] = 0.
if VERBOSE:
print(f"failed getting variance with {e}")
return results, misses
def recon_metrics(recon_sets, outpath, exp_name, name, epoch, VERBOSE):
misses = 0.
results = {
'fscores': [],
'iou_shape': [],
'param_dist_parts': [],
}
for prog, gt_prog, prog_ind in recon_sets:
bbox = getBBox(gt_prog)
gt_verts, gt_faces, gt_hscene = hier_execute(gt_prog, return_all=True)
gt_cubes = [[CuboidToParams(c) for c in scene] for scene in gt_hscene]
try:
verts, faces, hscene = hier_execute(prog, return_all=True)
cubes = [[CuboidToParams(c) for c in scene] for scene in hscene]
assert not torch.isnan(verts).any(), 'saw nan vert'
except Exception as e:
misses += 1.
if VERBOSE:
print(f"failed recon metrics for {prog_ind} with {e}")
continue
verts = verts.to(device)
gt_verts = gt_verts.to(device)
faces = faces.to(device)
gt_faces = gt_faces.to(device)
gt_objs = os.listdir(f"{outpath}/{exp_name}/objs/gt/")
if f"{prog_ind}.obj" not in gt_objs:
utils.writeObj(gt_verts, gt_faces,
f"{outpath}/{exp_name}/objs/gt/{prog_ind}.obj")
utils.writeHierProg(
gt_prog, f"{outpath}/{exp_name}/programs/gt/{prog_ind}.txt")
try:
utils.writeObj(
verts, faces,
f"{outpath}/{exp_name}/objs/{name}/{epoch}_{prog_ind}.obj")
utils.writeHierProg(
prog,
f"{outpath}/{exp_name}/programs/{name}/{epoch}_{prog_ind}.txt")
except Exception as e:
print(f"Failed writing prog/obj for {prog_ind} with {e}")
try:
fs = getFScore(verts, faces, gt_verts, gt_faces)
if fs is not None:
results['fscores'].append(fs)
except Exception as e:
if VERBOSE:
print(f"failed Fscore for {prog_ind} with {e}")
try:
siou = getShapeIoU(cubes, gt_cubes, bbox)
if siou is not None:
results['iou_shape'].append(siou)
except Exception as e:
if VERBOSE:
print(f"failed Shape Iou for {prog_ind} with {e}")
try:
pd = getParamDist(cubes, gt_cubes, bbox)
if pd is not None:
results['param_dist_parts'].append(pd)
except Exception as e:
if VERBOSE:
print(f"failed param dist for {prog_ind} with {e}")
for key in results:
if len(results[key]) > 0:
res = torch.tensor(results[key]).mean().item()
else:
res = 0.
results[key] = res
return results, misses
q += next_q
while len(q) > 0:
prog, hier_index, bbox_dims = q.pop(0)
next_q = rand_program(prog, num_cuboids, bbox_dims, hier_index)
if next_q is None:
prog.pop('prog')
prog.pop('children')
num_prev_children = 0
else:
q += next_q
num_prev_children = len(prog['children'])
num_cuboids = num_cuboids - num_prev_children + 2
# if not enough remaining cuboids to keep expanding
# pull all children out of q and make them leaves
if not num_cuboids > 2:
while len(q) > 0:
prog, hier_index, bbox_dims = q.pop(0)
prog.pop('prog')
prog.pop('children')
break
return hier_prog
for n in range(10):
prog = rand_hier_program()
verts, faces = hier_execute(prog)
writeObj(verts, faces, f'{n}.obj')
writeHierProg(prog, f"{n}.txt")
errors = 0.
gpfsv = 0.
gpcfsv = 0.
os.system(f'mkdir {outdir}')
for ind in inds:
count += 1.
try:
hier = jp.parseJsonToHier(ind, CATEGORY)
nshier = jp.parseJsonToHier(ind, CATEGORY, True)
gen.generate_program(hier)
pverts, pfaces = hier_execute(hier)
tverts, tfaces = get_gt_geom(nshier, False)
tsamps = utils.sample_surface(tfaces, tverts.unsqueeze(0), 10000)
try:
psamps = utils.sample_surface(pfaces, pverts.unsqueeze(0),
10000)
pfs = fscore.score(psamps.squeeze().T.unsqueeze(0),
tsamps.squeeze().T.unsqueeze(0))
except Exception:
pfs = 0.
def fit(prog_path, obj_path, out_path):
progs = os.listdir(prog_path)
objs = os.listdir(obj_path)
fitted_progs = []
for i, prg in enumerate(progs):
print(f"fitting program {i}")
sa = ShapeAssembly()
p_no_e = prg.split("_")[1]
index = int(p_no_e.split(".")[0])
# should be shape N x 3
tverts, tfaces = loadObj(f"{obj_path}/{index}.obj")
tverts = torch.tensor(tverts)
tfaces = torch.tensor(tfaces).long()
out_file = f"{out_path}/{index}"
with open(f"{prog_path}/{prg}") as file:
lines = file.readlines()
hier, param_dict, param_list = sa.make_hier_param_dict(lines)
opt = torch.optim.Adam(param_list, 0.001)
start = torch.cat(param_list).clone()
for iter in range(400):
verts, faces = sa.diff_run(hier, param_dict)
samps = sample_surface(faces, verts.unsqueeze(0), 10000)
tsamps = sample_surface(tfaces, tverts.unsqueeze(0), 10000)
closs = cham_loss(samps.squeeze().T.unsqueeze(0).cuda(),
tsamps.squeeze().T.unsqueeze(0).cuda(), 0.0)
ploss = (torch.cat(param_list) - start).abs().sum()
loss = closs + ploss.cuda() * 0.001
opt.zero_grad()
loss.backward()
opt.step()
# # prevent cuboids from having 0 dimensions
new_param_dict = {}
for p in param_dict:
new_p = []
for param in param_dict[p]:
if param[0] == "Cuboid":
new_attrs = []
for attr in param[1]:
if torch.is_tensor(attr):
new_attr = torch.clamp(attr, min=0.01).detach()
new_attrs.append(new_attr)
else:
new_attrs.append(attr)
new_p.append((param[0], new_attrs))
else:
new_p.append(param)
new_param_dict[p] = new_p
sa.fill_hier(hier, new_param_dict)
verts, faces = hier_execute(hier)
writeObj(verts, faces, out_file + '.obj')
writeHierProg(hier, out_file + '.txt')
fitted_progs.append((hier, index))
return fitted_progs