I run yolov4 on Jetson NX,but can not get correct bbox use darknet2onnx.py to convert yolov4.cfg and yolov4.weight. then convert to yolov4.trt when run demo_trt.py yolov4.trt dog.jpg 416 416 get below result /----------------------------------------- Reading engine from file yolov4-416.trt Shape of the network input: (1, 3, 416, 416) Length of inputs: 1 Len of outputs: 9

TRT inference time: 0.132841

 get_region_boxes : 0.002494
              nms : 0.007532

post process total : 0.010035

truck: 0.980625 dog: 0.999999 bicycle: 1.000000 /---------------------------------- but no bbox in the predictions_trt.jpg I print the <trt_outputs> check the in demo_trt.pyp line 151 get below value

/------------------------------------------- [array([1.0889691e-04, 3.5633548e-05, 2.2827997e-05, ..., 3.6400868e-04, 2.5315638e-04, 3.8831109e-05], dtype=float32), array([7.9498291e-03, 1.0000000e+01, 1.8692017e-02, ..., 1.0000000e+01, 4.8004150e-02, 3.5705566e-02], dtype=float32), array([4.1845703e-01, 1.3954163e-02, 8.3496094e-02, ..., 6.1416626e-03, 8.0537796e-04, 1.5139580e-04], dtype=float32), array([4.0929112e-06, 8.2035322e-06, 6.0963603e-06, ..., 4.5659704e-06, 2.2603242e-06, 8.5792357e-07], dtype=float32), array([ 0.02755033, 10. , 0.07742122, ..., 10. , 0.10579979, 0.13517565], dtype=float32), array([3.6437038e-01, 1.7008580e-02, 5.5334978e-02, ..., 7.9847518e-03, 2.9890699e-04, 2.4877247e-04], dtype=float32), array([1.04296953e-06, 2.07851699e-05, 3.73435687e-05, 3.64785374e-05, ...................... , array([0.02295261, 0.00861033, 0.12978163, ..., 0.00506801, 0.00085359, 0.00099795], dtype=float32), array([ 0.06552535, 10. , 0.24504207, ..., 10. , 0.5348558 , 0.55544066], dtype=float32)] //------------------------------------------------------------------------------- I got some wrong value==> 10. ..10. But it is really strange in my PC RTX2070 TRV7.0.0.11 is ok ,no problem. I have no idea about it.

Traceback (most recent call last): File "train.py", line 635, in device=device, ) File "train.py", line 425, in train evaluator = evaluate(eval_model, val_loader, config, device) File "/root/anaconda2/envs/pytorch_1_5/lib/python3.6/site-packages/torch/autograd/grad_mode.py", line 15, in decorate_context return func(*args, **kwargs) File "train.py", line 493, in evaluate for img, target, (boxes, confs) in zip(images, targets, outputs): ValueError: too many values to unpack (expected 2)

Is there any useful solution

我自行修改了一个版本 [[[0.5532519, 0.6153389, 0.8914345, 0.82043, 0.96487856, 0.96487856, 0]]] person: 0.964879 Shape of the network input: (5, 3, 320, 320) Length of inputs: 1 [TensorRT] WARNING: Explicit batch network detected and batch size specified, use enqueue without batch size instead. Len of outputs: 1 可是输出还是为1批次的。 顺便希望加微信群，我可以尝试完成deepstream的移植工作，希望进群讨论

您好，是训练好pth模型后，将pth模型转为onnx，然后使用demo_onnx.py进行onnx推理，但onnx的输出全是： [[nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], ..., [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan], [nan, nan, nan, ..., nan, nan, nan]],

    [[nan, nan, nan, ..., nan, nan, nan],

....

请问是要修改哪里吗？？我转出的onnx完全没法使用

RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.FloatTensor [4, 3, 19, 19, 85]], which is output 0 of AsStridedBackward, is at version 6; expected version 3 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True). 我训练自己的数据集遇到这个问题，之前重来没有遇到过，请问有解决办法吗？

I added torch weights support for demo.py by adding a new argument "-torch" to avoid issues like this when using demo.py.

I also fixed the false warning whenever any yolo cfg file is initialized by darknet2pytorch.Darknet when it parses the Linear activation of the convolutional blocks as I mentioned here.

I am loading a pretrained model with these cfg settings:

 Cfg.use_darknet_cfg = True
 Cfg.cfgfile = os.path.join(_BASE_DIR, 'cfg', 'yolov4-custom.cfg')

I also edited the classes settings in 'yolov4-custom.cfg' from 80 to 3 ( as in my dataset). Finally this is my script to load the model and evaluate on a give dataset:

 ## Darknet

 model = Darknet(Cfg.cfgfile)
## Custom YOLO
#model = Yolov4(yolov4conv137weight=None, n_classes=3, inference=True)
pretrained_dict = torch.load(load_path, map_location=device)
model.load_state_dict(pretrained_dict)
model.to(device = device)
 
"Build datasets"
cfg = get_args(**Cfg)
train_dataset = Yolo_dataset(train_label, Cfg)
val_dataset = Yolo_dataset(val_label, Cfg)
test_dataset = Yolo_dataset(test_label, Cfg)

"Evaluate train set"
 train_loader = DataLoader(train_dataset, batch_size= 4, shuffle=True,
                              num_workers=0, pin_memory=True, drop_last=True, collate_fn = collate)
criterion = Yolo_loss(device = device, batch = len(train_dataset),n_classes = 3)

 with torch.no_grad():
   for i, batch in enumerate(train_loader):
       images = batch[0]
       bboxes = batch[1]
       images = images.to(device= device, dtype=torch.float32)
       bboxes = bboxes.to(device = device)
       model.eval()
       bboxes_pred = model(images)
       train_loss, train_loss_xy, train_loss_wh, train_loss_obj, train_loss_cls, train_loss_l2 = criterion(bboxes_pred, bboxes)
       print(train_loss)

I don't understand why the shape of bboxes_pred is [4, 22743,7] ?? Can someone help me out?

Traceback (most recent call last): File "models.py", line 462, in boxes = do_detect(model, sized, 0.5, n_classes,0.4, use_cuda) File "/home/ubuntu/pytorch-YOLOv4/tool/torch_utils.py", line 247, in do_detect output[-1].append(boxes_and_confs[i][1].cpu().detach().numpy()) IndexError: index 1 is out of bounds for dimension 0 with size 1 any help appreciated!

I tried to inference on multiple images in a loop. It copies output of the first image into the output of all others!! Is it a bug? Or there is a different way to do multi-inference?

Hi, I train to train the model on my dataset, other parts work fine but when it starts training, it raises an error, "shape '[4, 3, 8, 76, 76]' is invalid for the input of size 5891520. Does anyone know what might cause this error?

Whwn I finish training the model with yolov4,and the result is look pretty good. I use it to test my test datasets, the accuracy is nice but sometimes will happend double recognize. What's the reason to cause the problem?

When I introduce random noise to the input loss.item() decreases. Part of the code for training is included here.

  for epoch in range(epochs):
        # model.train()
        epoch_loss = 0
        epoch_step = 0

        with tqdm(total=n_train, desc=f'Epoch {epoch + 1}/{epochs}', unit='img', ncols=50) as pbar:
            for i, batch in enumerate(train_loader):
                global_step += 1
                epoch_step += 1
                images = batch[0]
                bboxes = batch[1]
                paths = batch[2]

                images = images.to(device=device, dtype=torch.float32)
                images.requires_grad = True
                bboxes = bboxes.to(device=device)
                bboxes_pred = model(images)

                loss, loss_xy, loss_wh, loss_obj, loss_cls, loss_l2 = criterion(
                    bboxes_pred, bboxes)

                print("Without Noise - ", loss.item(), loss_xy.item(), loss_wh.item(),
                      loss_obj.item(), loss_cls.item(), loss_l2.item())

                grad = torch.autograd.grad(loss, [images], create_graph=True,
                                           retain_graph=True, allow_unused=True)[0]

                # loss.backward()

                noise = torch.randn(images.size())
                imgs = images.detach().cpu() + noise

                imgs = imgs.to(device=device, dtype=torch.float32)

                bboxes_pred = model(imgs)

                loss, loss_xy, loss_wh, loss_obj, loss_cls, loss_l2 = criterion(
                    bboxes_pred, bboxes)

                print("With Noise - ", loss.item(), loss_xy.item(), loss_wh.item(),
                      loss_obj.item(), loss_cls.item(), loss_l2.item())

Output - Without Noise - 20.36496353149414 0.0 0.0 20.36496353149414 0.0 6.040985584259033 With Noise - 1.169505000114441 0.0 0.0 1.169505000114441 0.0 0.05301150307059288 --------- looping again Without Noise - 25.24418830871582 0.0 0.0 25.24418830871582 0.0 7.514528274536133 With Noise - 1.8068726062774658 0.0 0.0 1.8068726062774658 0.0 0.12245824933052063 --------- looping again Without Noise - 1.1534618139266968 0.0 0.0 1.1534618139266968 0.0 0.2820863425731659 With Noise - 0.8495630025863647 0.0 0.0 0.8495630025863647 0.0 0.034061141312122345 --------- looping again Without Noise - 2754.517333984375 3.337832450866699 2730.53369140625 18.649364471435547 1.9963617324829102 5465.08642578125 With Noise - 2763.16845703125 2.5691161155700684 2741.94189453125 13.955108642578125 4.702309608459473 5486.26513671875

Bumps pillow from 7.1.2 to 9.3.0.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

File "models.py", line 509, in plot_boxes_cv2(img, boxes[0], 'predictions.jpg', class_names) File "/home/duys/workspace/pytorch-YOLOv4/tool/utils.py", line 140, in plot_boxes_cv2 cv2.rectangle(img, (x1,y1), (np.float32(c3[0]), np.float32(c3[1])), rgb, -1) cv2.error: OpenCV(4.6.0) :-1: error: (-5:Bad argument) in function 'rectangle'

Overload resolution failed:

• Can't parse 'pt2'. Sequence item with index 0 has a wrong type
• Can't parse 'pt2'. Sequence item with index 0 has a wrong type
• Can't parse 'rec'. Expected sequence length 4, got 2
• Can't parse 'rec'. Expected sequence length 4, got 2

do_detect() has an extra argument. Comparing to previous version used to be boxes = do_detect(model, sized, 0.0, 0.4, use_cuda) but now it is boxes = do_detect(model, sized, 0.0, 80, 0.4, use_cuda) Is this a typo? do_detect() wasn't changed to accommodate an extra argument.