Nguyen Van Thanh
Nguyen Van Thanh
@junyanz I meet the same problem. I set `--display_id 0`. I run on Windows. Do you have any suggesstion?
You can look at his code. This part : LABEL_NAMES = np.asarray([ 'background', 'aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep',...
Yes, I still have the issue after setting `--display_id 0` . I stopped debugging and change to Linux. Tks anw.
Hi @hsauod , did you try add --vis_fast ? It seems stuck at rendering (sometimes)
Hi, i'm confused too. Hi @lindatan90 From what i read, [train.py line 104](https://github.com/aimagelab/show-control-and-tell/blob/master/train.py), captions shape should be (bs, 20). I printed. For detection: detections.shape: torch.Size([bs, 100, 2048]: first dim is...
You might want to read [model_file](https://github.com/aimagelab/show-control-and-tell/blob/master/models/controllable_captioning.py) We extract image descriptor by get mask of detection tensor, get average of feature, concatenate [embedded word, image descriptor] to create inputs. seqs[1] =...
For Sinkhorn operator training, you could read their paper and write your own code.
This is mean there is (at least) one img in your dataset doesn't contain any bbox.
@rttariverdi67 Did you check your dataset? Is the image creating this error readable? How many channel does it have?
You can temporary fix by add these 2 lines on top of files: ``` import sys sys.path.append('.') ```