yuxumin / core Goto Github PK

Hi, Thanks for this amazing work. I ran your code on AQA-7 dataset (with exactly same settings) but the results were a bit lower than the reported results in the paper. I was wondering if you could let me know whether this is the last version of your code or I need to use different settings to get the same results as you reported in the paper. Thanks
In the following table CoRe* is our reproduced results on AQA-7.

Hi,
Thanks for your amazing work on AQA tasks. Since I am working on the skill assessment task, may I ask when you will release your training and test code for the JIGSAW dataset? I look forward to your reply!

Hi could you please tell me where did you get this library

from torchvideotransforms import video_transforms, volume_transforms

I installed pytorchvideo library but looks like its not that. Thanks.

    total_video = torch.cat((target,exemplar),0)  # 2N C H W
    start_idx = [0,10,20,30,40,50,60,70,80,86]
    video_pack = torch.cat([total_video[:, :, i : i + 16] for i in start_idx])  # 10*2N, c, 16, h, w
    total_feature = self.backbone(video_pack).reshape(10,len(total_video),-1).transpose(0,1)  # 2N * 10 * 1024
    
    
    comment says video_pack shapes like 10*2N, c, 16, h, w 
    
    but when i try it ，it shapes like  10*2N, c, 16,  w 

is there any wrong in my codes？

video = []
for i in range(100):
image = Image.open(os.path.join(image_path, 'image_%06d.jpg' % ( i + 1 )))
video.append(image)
trans, _ = get_video_trans()
t = trans(video).float()
print(t.shape)
t = torch.cat((t,t),0) # 2N C H W
print(t.shape)
start_idx = [0,10,20,30,40,50,60,70,80,86]
video_pack = torch.cat([t[:, :, i : i + 16] for i in start_idx]) # 10*2N, c, 16, h, w
print(video_pack.shape)

which print
torch.Size([3, 100, 224, 224])
torch.Size([6, 100, 224, 224])
torch.Size([60, 100, 16, 224])

After reading your code, I think it equals a multi-classification and regression problem. Why did not you directly get the final classification results but by 2 to 4 to 8 to 16? Does this progressive method help a lot?