import os,io,shutil
import geometry
import wandb
from matplotlib import cm
import cv2
from tqdm import tqdm
import torchvision
import time
from torchvision.utils import make_grid,draw_keypoints
import torch.nn.functional as F
import kornia
import numpy as np
import torch
import flow_vis
import flow_vis_torch
import matplotlib.pyplot as plt
from einops import rearrange, repeat
import models
import piqa
import imageio
from PIL import Image
#import splines.quaternion
#from torchcubicspline import (natural_cubic_spline_coeffs, NaturalCubicSpline)
from scipy import spatial
import plotly.express as px
import plotly.graph_objects as go
from collections import defaultdict
import geometry

ch_fst = lambda src,x=None:rearrange(src,"... (x y) c -> ... c x y",x=int(src.size(-2)**(.5)) if x is None else x)
ch_sec = lambda x: rearrange(x,"... c x y -> ... (x y) c")

def wandb_summary(loss, model_output, model_input, ground_truth, resolution,prefix="",suffix="",step=0,losses_agg=[]):
    model_output,model_input,ground_truth = [{k:(v[:1] if len(v.shape) else v) for k,v in x.items()} for x in (model_output,model_input,ground_truth)]

    wandb_out = {}

    # Log images
    for k,v in {k:v for k,v in model_input.items() if "img" in k}.items(): wandb_out["ref/"+k]= make_grid(v.flatten(0,1))

    wandb_out["ref/pointmap"]= make_grid(model_input["pointmap"],normalize=True)
    wandb_out["est/pointmap"]= make_grid(model_output["pointmap"],normalize=True)

    seg_cmap=plt.cm.get_cmap('tab20', model_input["segs"].size(1))
    seg_colors = [torch.tensor(seg_cmap(i))[:3] for i in range(model_input["segs"].size(1))]
    seg_img = ch_sec(torch.ones_like(model_input["pointmap"]))[0]
    for seg,seg_color in zip(model_input["segs"].unbind(1),seg_colors): seg_img[ch_sec(seg).squeeze(1)]=seg_color.cuda()
    wandb_out["est/seg"]= make_grid(ch_fst(seg_img,model_input["pointmap"].size(-2)),normalize=False)

    #print("Solving for transfs")
    #transfs_recovered = torch.stack([geometry.procrustes(ch_sec(model_input["canon_pointmap"])[0][seg.flatten()][None].float(),ch_sec(model_input["pointmap"])[0][seg.flatten()][None].float())[1] 
    #                                                                        for seg in model_input["segs"][:,1:].unbind(1)])
    # Solve for transfs per object
    seg=model_input["segs"][:,1]
    est_transf = geometry.procrustes(ch_sec(model_input["canon_pointmap"])[0][seg.flatten()][None].float(),ch_sec(model_output["pointmap"])[0][seg.flatten()][None].float())[1] 

    # Visualize transfs as unit tri deformations
    unit_tri = torch.tensor([ [0, 0, 0,1],  [1, 0, 0,1],  [0, 1, 0,1],  ]).float().numpy()
    model_output["pred_tris"]=geometry.transf_to_tris(est_transf)[0]
    model_input["transf_tris"]=geometry.transf_to_tris(model_input["obj_transfs"][:,1].float())[0]
    wandb.log({"tri_vis_metric_"+prefix: (model_output["pred_tris"]-model_input["transf_tris"]).square().mean()}, step=step)
    pred_tri = model_output["pred_tris"].detach().cpu().numpy()[0]
    gt_tri = model_input["transf_tris"].detach().cpu().numpy()[0]
    fig = plt.figure(figsize=(10, 10))
    ax = fig.add_subplot(111, projection='3d')
    # add label here
    ax.plot_trisurf(unit_tri[:, 0], unit_tri[:, 1], unit_tri[:, 2], color="red", alpha=0.5)
    ax.plot_trisurf(pred_tri[:, 0], pred_tri[:, 1], pred_tri[:, 2], color=seg_colors[0].numpy() if 0 else "green", alpha=0.5)
    ax.plot_trisurf(gt_tri[:, 0], gt_tri[:, 1], gt_tri[:, 2],       color=seg_colors[0].numpy() if 0 else "blue", alpha=0.5)
    try:
        plt.savefig("output/img/tmp.png");plt.close();
        img_arr=plt.imread("output/img/tmp.png")
        wandb_out["est/tri_obj1"] = torch.from_numpy(img_arr[...,:3]).permute(2,0,1)
        print("logging images",len(wandb_out))
    except: print("failed saving/logging 3d plot image")
    plt.close()

    if 0:
        for k,v in wandb_out.items(): print(k,v.max(),v.min())
        for k,v in wandb_out.items():
            print(k,v.shape)
            plt.imsave("output/img/%s.png"%k,v.float().permute(1,2,0).detach().cpu().numpy().clip(0,1));
        print("saving locally")
        zz

    for k,v in wandb_out.items():print(k,v.shape)
    wandb.log({prefix+k:wandb.Image(v.permute(1, 2, 0).float().detach().clip(0,1).cpu().numpy()) for k,v in wandb_out.items()})
    print("done logging images")
    return wandb_out