VISAM/submit_dance.py

# ------------------------------------------------------------------------
# Copyright (c) 2022 megvii-research. All Rights Reserved.
# ------------------------------------------------------------------------
# Modified from Deformable DETR (https://github.com/fundamentalvision/Deformable-DETR)
# Copyright (c) 2020 SenseTime. All Rights Reserved.
# ------------------------------------------------------------------------
# Modified from DETR (https://github.com/facebookresearch/detr)
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
# ------------------------------------------------------------------------

from copy import deepcopy
import json

import os
import argparse
import torchvision.transforms.functional as F
import torch
import cv2
import numpy as np
from tqdm import tqdm
from pathlib import Path
from models import build_model
from util.tool import load_model
from main import get_args_parser

from models.structures import Instances
from torch.utils.data import Dataset, DataLoader


import sys
sys.path.append('thirdparty/segment-anything')
from segment_anything import build_sam, SamPredictor 

class Colors:
    # Ultralytics color palette https://ultralytics.com/
    def __init__(self):
        # hex = matplotlib.colors.TABLEAU_COLORS.values()
        hexs = ('FF3838', 'FF9D97', 'FF701F', 'FFB21D', 'CFD231', '48F90A', '92CC17', '3DDB86', '1A9334', '00D4BB',
                '2C99A8', '00C2FF', '344593', '6473FF', '0018EC', '8438FF', '520085', 'CB38FF', 'FF95C8', 'FF37C7')
        self.palette = [self.hex2rgb(f'#{c}') for c in hexs]
        self.n = len(self.palette)

    def __call__(self, i, bgr=False):
        c = self.palette[int(i) % self.n]
        return (c[2], c[1], c[0]) if bgr else c

    @staticmethod
    def hex2rgb(h):  # rgb order (PIL)
        return tuple(int(h[1 + i:1 + i + 2], 16) for i in (0, 2, 4))


colors = Colors()  # create instance for 'from utils.plots import colors'


class ListImgDataset(Dataset):
    def __init__(self, mot_path, img_list, det_db) -> None:
        super().__init__()
        self.mot_path = mot_path
        self.img_list = img_list
        self.det_db = det_db

        '''
        common settings
        '''
        self.img_height = 800
        self.img_width = 1536
        self.mean = [0.485, 0.456, 0.406]
        self.std = [0.229, 0.224, 0.225]

    def load_img_from_file(self, f_path):
        cur_img = cv2.imread(os.path.join(self.mot_path, f_path))
        assert cur_img is not None, f_path
        cur_img = cv2.cvtColor(cur_img, cv2.COLOR_BGR2RGB)
        proposals = []
        im_h, im_w = cur_img.shape[:2]
        for line in self.det_db[f_path[:-4] + '.txt']:
            l, t, w, h, s = list(map(float, line.split(',')))
            proposals.append([(l + w / 2) / im_w,
                                (t + h / 2) / im_h,
                                w / im_w,
                                h / im_h,
                                s])
        return cur_img, torch.as_tensor(proposals).reshape(-1, 5)

    def init_img(self, img, proposals):
        ori_img = img.copy()
        self.seq_h, self.seq_w = img.shape[:2]
        scale = self.img_height / min(self.seq_h, self.seq_w)
        if max(self.seq_h, self.seq_w) * scale > self.img_width:
            scale = self.img_width / max(self.seq_h, self.seq_w)
        target_h = int(self.seq_h * scale)
        target_w = int(self.seq_w * scale)
        img = cv2.resize(img, (target_w, target_h))
        img = F.normalize(F.to_tensor(img), self.mean, self.std)
        img = img.unsqueeze(0)
        return img, ori_img, proposals

    def __len__(self):
        return len(self.img_list)
    
    def __getitem__(self, index):
        img, proposals = self.load_img_from_file(self.img_list[index])
        return self.init_img(img, proposals)


class Detector(object):
    def __init__(self, args, model, vid):
        self.args = args
        self.detr = model

        self.vid = vid
        self.seq_num = os.path.basename(vid)
        img_list = os.listdir(os.path.join(self.args.mot_path, vid, 'img1'))
        img_list = [os.path.join(vid, 'img1', i) for i in img_list if 'jpg' in i]

        self.img_list = sorted(img_list)
        self.img_len = len(self.img_list)

        self.predict_path = os.path.join(self.args.output_dir, args.exp_name)
        os.makedirs(self.predict_path, exist_ok=True)

        
        self.sam_predictor = SamPredictor(build_sam(checkpoint="thirdparty/segment-anything/sam_vit_h_4b8939.pth"))
        _ = self.sam_predictor.model.to(device='cuda')
        
        fps = 25
        size = (1920, 1080) 
        self.videowriter = cv2.VideoWriter('visam.avi', cv2.VideoWriter_fourcc('M','J','P','G'), fps, size)

    @staticmethod
    def filter_dt_by_score(dt_instances: Instances, prob_threshold: float) -> Instances:
        keep = dt_instances.scores > prob_threshold
        keep &= dt_instances.obj_idxes >= 0
        return dt_instances[keep]

    @staticmethod
    def filter_dt_by_area(dt_instances: Instances, area_threshold: float) -> Instances:
        wh = dt_instances.boxes[:, 2:4] - dt_instances.boxes[:, 0:2]
        areas = wh[:, 0] * wh[:, 1]
        keep = areas > area_threshold
        return dt_instances[keep]

    def detect(self, prob_threshold=0.6, area_threshold=100, vis=False):
        total_dts = 0
        total_occlusion_dts = 0

        track_instances = None
        with open(os.path.join(self.args.mot_path, 'DanceTrack', self.args.det_db)) as f:
            det_db = json.load(f)
        loader = DataLoader(ListImgDataset(self.args.mot_path, self.img_list, det_db), 1, num_workers=2)
        lines = []
        for i, data in enumerate(tqdm(loader)):
            cur_img, ori_img, proposals = [d[0] for d in data]
            cur_img, proposals = cur_img.cuda(), proposals.cuda()

            # track_instances = None
            if track_instances is not None:
                track_instances.remove('boxes')
                track_instances.remove('labels')
            seq_h, seq_w, _ = ori_img.shape

            res = self.detr.inference_single_image(cur_img, (seq_h, seq_w), track_instances, proposals)
            track_instances = res['track_instances']

            dt_instances = deepcopy(track_instances)

            # filter det instances by score.
            dt_instances = self.filter_dt_by_score(dt_instances, prob_threshold)
            dt_instances = self.filter_dt_by_area(dt_instances, area_threshold)

            total_dts += len(dt_instances)

            bbox_xyxy = dt_instances.boxes.tolist()
            identities = dt_instances.obj_idxes.tolist()


            masks_all = []
            self.sam_predictor.set_image(ori_img.to(torch.device('cpu')).numpy().copy())
            
            for bbox, id in zip(np.array(bbox_xyxy), identities):
                masks, iou_predictions, low_res_masks = self.sam_predictor.predict(box=bbox)
                index_max = iou_predictions.argsort()[0]
                masks = np.concatenate([masks[index_max:(index_max+1)], masks[index_max:(index_max+1)], masks[index_max:(index_max+1)]], axis=0)
                masks = masks.astype(np.int32)*np.array(colors(id))[:, None, None]
                masks_all.append(masks)
            
            self.sam_predictor.reset_image()
            if len(masks_all):
                masks_sum = masks_all[0].copy()
                for m in masks_all[1:]:
                    masks_sum += m
            else:
                masks_sum = np.zeros_like(img).transpose(2, 0, 1)

            img = ori_img.to(torch.device('cpu')).numpy().copy()[..., ::-1]
            img = (img * 0.5 + (masks_sum.transpose(1,2,0) * 30) %128).astype(np.uint8)
            for bbox in bbox_xyxy:
                cv2.rectangle(img, (int(bbox[0]), int(bbox[1])), (int(bbox[2]), int(bbox[3])), (0,0,255), thickness=3)
            self.videowriter.write(img)


            save_format = '{frame},{id},{x1:.2f},{y1:.2f},{w:.2f},{h:.2f},1,-1,-1,-1\n'
            for xyxy, track_id in zip(bbox_xyxy, identities):
                if track_id < 0 or track_id is None:
                    continue
                x1, y1, x2, y2 = xyxy
                w, h = x2 - x1, y2 - y1
                lines.append(save_format.format(frame=i + 1, id=track_id, x1=x1, y1=y1, w=w, h=h))
        with open(os.path.join(self.predict_path, f'{self.seq_num}.txt'), 'w') as f:
            f.writelines(lines)
        print("totally {} dts {} occlusion dts".format(total_dts, total_occlusion_dts))

class RuntimeTrackerBase(object):
    def __init__(self, score_thresh=0.6, filter_score_thresh=0.5, miss_tolerance=10):
        self.score_thresh = score_thresh
        self.filter_score_thresh = filter_score_thresh
        self.miss_tolerance = miss_tolerance
        self.max_obj_id = 0

    def clear(self):
        self.max_obj_id = 0

    def update(self, track_instances: Instances):
        device = track_instances.obj_idxes.device

        track_instances.disappear_time[track_instances.scores >= self.score_thresh] = 0
        new_obj = (track_instances.obj_idxes == -1) & (track_instances.scores >= self.score_thresh)
        disappeared_obj = (track_instances.obj_idxes >= 0) & (track_instances.scores < self.filter_score_thresh)
        num_new_objs = new_obj.sum().item()

        track_instances.obj_idxes[new_obj] = self.max_obj_id + torch.arange(num_new_objs, device=device)
        self.max_obj_id += num_new_objs

        track_instances.disappear_time[disappeared_obj] += 1
        to_del = disappeared_obj & (track_instances.disappear_time >= self.miss_tolerance)
        track_instances.obj_idxes[to_del] = -1


if __name__ == '__main__':

    parser = argparse.ArgumentParser('DETR training and evaluation script', parents=[get_args_parser()])
    parser.add_argument('--score_threshold', default=0.5, type=float)
    parser.add_argument('--update_score_threshold', default=0.5, type=float)
    parser.add_argument('--miss_tolerance', default=20, type=int)
    args = parser.parse_args()
    if args.output_dir:
        Path(args.output_dir).mkdir(parents=True, exist_ok=True)

    # load model and weights
    detr, _, _ = build_model(args)
    detr.track_embed.score_thr = args.update_score_threshold
    detr.track_base = RuntimeTrackerBase(args.score_threshold, args.score_threshold, args.miss_tolerance)
    checkpoint = torch.load(args.resume, map_location='cpu')
    detr = load_model(detr, args.resume)
    detr.eval()
    detr = detr.cuda()

    # '''for MOT17 submit''' 
    sub_dir = 'DanceTrack/test'
    seq_nums = os.listdir(os.path.join(args.mot_path, sub_dir))
    if 'seqmap' in seq_nums:
        seq_nums.remove('seqmap')
    vids = [os.path.join(sub_dir, seq) for seq in seq_nums]

    rank = int(os.environ.get('RLAUNCH_REPLICA', '0'))
    ws = int(os.environ.get('RLAUNCH_REPLICA_TOTAL', '1'))
    vids = vids[rank::ws]

    for vid in vids:
        det = Detector(args, model=detr, vid=vid)
        det.detect(args.score_threshold)