calPCKH_fashion.py

import tqdm
import pandas as pd
import json

MISSING_VALUE = -1

PARTS_SEL = [0, 1, 14, 15, 16, 17]

target_annotation = '../datasets/deepfashion/label/fasion-annotation-test.csv'
pred_annotation = '../checkpoints/PoseTransfer_deepfashion/output_pckh.csv'


'''
  hz: head size
  alpha: norm factor
  px, py: predict coords
  tx, ty: target coords
'''
def isRight(px, py, tx, ty, hz, alpha):
    if px == -1 or py == -1 or tx == -1 or ty == -1:
        return 0

    if abs(px - tx) < hz[0] * alpha and abs(py - ty) < hz[1] * alpha:
        return 1
    else:
        return 0


def how_many_right_seq(px, py, tx, ty, hz, alpha):
    nRight = 0
    for i in range(len(px)):
        nRight = nRight + isRight(px[i], py[i], tx[i], ty[i], hz, alpha)

    return nRight


def ValidPoints(tx):
    nValid = 0
    for item in tx:
        if item != -1:
            nValid = nValid + 1
    return nValid


def get_head_wh(x_coords, y_coords):
    final_w, final_h = -1, -1
    component_count = 0
    save_componets = []
    for component in PARTS_SEL:
        if x_coords[component] == MISSING_VALUE or y_coords[component] == MISSING_VALUE:
            continue
        else:
            component_count += 1
            save_componets.append([x_coords[component], y_coords[component]])
    if component_count >= 2:
        x_cords = []
        y_cords = []
        for component in save_componets:
            x_cords.append(component[0])
            y_cords.append(component[1])
        xmin = min(x_cords)
        xmax = max(x_cords)
        ymin = min(y_cords)
        ymax = max(y_cords)
        final_w = xmax - xmin
        final_h = ymax - ymin
    return final_w, final_h


tAnno = pd.read_csv(target_annotation, sep=':')
pAnno = pd.read_csv(pred_annotation, sep=':')

pRows = pAnno.shape[0]

nAll = 0
nCorrect = 0
alpha = 0.5
for i in tqdm.tqdm(range(pRows)):
    pValues = pAnno.iloc[i].values
    pname = pValues[0]
    pycords = json.loads(pValues[1])  # list of numbers
    pxcords = json.loads(pValues[2])

    tname = pname

    ####
    tname = tname.replace('.jpg_vis.jpg','.jpg')
    tname = tname.replace('.jpg___', '_')
    if tname.count('_')==5:
        ns = tname.split('_')
        tname = ns[0]+ns[1]+'_'+ns[2]+'_'+ns[3]+ns[4]+'_'+ns[5]
    tname = tname.replace('fashion', 'fasion')
    tValues = tAnno.query('name == "%s"' % (tname)).values[0]
    tycords = json.loads(tValues[1])  # list of numbers
    txcords = json.loads(tValues[2])


    xBox, yBox = get_head_wh(txcords, tycords)
    if xBox == -1 or yBox == -1:
        continue

    head_size = (xBox, yBox)
    nAll = nAll + ValidPoints(tycords)
    nCorrect = nCorrect + how_many_right_seq(pxcords, pycords, txcords, tycords, head_size, alpha)

print('%d/%d %f' % (nCorrect, nAll, nCorrect * 1.0 / nAll))