accuracy not as good as YOWO-Plus's

[luyuanmeng]

This is the result after 5 eopch of iteration，The video's accuracy is not there. But frame's accuracy is about the same as YOWO-Plus，So I'm a little puzzled and hope to get your answer
V-mAP @ 0.05 IoU:
--Per AP: [0.11, 0.05, 65.67, 97.17, 83.52, 100.0, 79.01, 100.0, 99.94, 96.08, 44.86, 93.51, 92.12, 100.0, 22.86, 97.59, 90.93, 96.55, 94.83, 63.46, 74.07, 48.76, 82.94, 88.3]
--mAP: 75.51

V-mAP @ 0.1 IoU:
--Per AP: [0.06, 0.0, 65.67, 94.22, 83.52, 100.0, 79.01, 100.0, 99.94, 96.08, 44.86, 93.51, 92.12, 100.0, 22.86, 97.59, 90.93, 96.55, 94.83, 63.46, 71.43, 48.76, 82.94, 88.3]
--mAP: 75.28

V-mAP @ 0.2 IoU:
--Per AP: [0.03, 0.0, 61.42, 91.4, 60.2, 100.0, 79.01, 100.0, 91.87, 96.08, 44.86, 90.6, 92.12, 100.0, 20.97, 97.59, 90.93, 96.55, 94.83, 63.46, 43.21, 47.51, 51.2, 88.3]
--mAP: 70.92

V-mAP @ 0.3 IoU:
--Per AP: [0.0, 0.0, 61.42, 72.45, 12.11, 88.2, 74.27, 100.0, 91.87, 93.52, 44.86, 88.11, 92.12, 100.0, 15.16, 97.59, 90.39, 96.55, 94.83, 51.36, 23.91, 45.08, 5.4, 83.68]
--mAP: 63.45

V-mAP @ 0.5 IoU:
--Per AP: [0.0, 0.0, 56.11, 33.23, 2.35, 44.21, 54.56, 100.0, 64.13, 91.44, 42.2, 71.48, 65.27, 100.0, 2.76, 97.59, 90.39, 89.4, 91.42, 45.01, 0.97, 20.9, 0.0, 66.03]
--mAP: 51.23

V-mAP @ 0.75 IoU:
--Per AP: [0.0, 0.0, 33.41, 0.0, 0.0, 3.7, 7.08, 75.5, 8.17, 35.82, 7.45, 31.58, 1.01, 61.67, 0.4, 78.97, 48.05, 26.43, 83.1, 13.05, 0.0, 0.64, 0.0, 29.94]
--mAP: 22.75

AP: 81.57% (1)
AP: 96.60% (10)
AP: 80.04% (11)
AP: 67.44% (12)
AP: 78.58% (13)
AP: 97.21% (14)
AP: 91.69% (15)
AP: 91.55% (16)
AP: 76.06% (17)
AP: 90.41% (18)
AP: 97.96% (19)
AP: 73.44% (2)
AP: 92.93% (20)
AP: 86.83% (21)
AP: 78.60% (22)
AP: 75.65% (23)
AP: 90.88% (24)
AP: 86.96% (3)
AP: 79.17% (4)
AP: 71.25% (5)
AP: 95.83% (6)
AP: 92.93% (7)
AP: 93.58% (8)
AP: 97.20% (9)

mAP: 86.01%

[yjh0410]

@luyuanmeng It seems that the model you trained achieves a better results.

[luyuanmeng]

But the video map is low compared to your results，I don't know why

…

---Original--- From: "Jianhua ***@***.***> Date: Mon, Nov 7, 2022 08:52 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [yjh0410/PyTorch_YOWO] accuracy not as good as YOWO-Plus's (Issue#6) @luyuanmeng It seems that the model you trained achieves a better results. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[yjh0410]

@luyuanmeng Your V-mAP@0.5 is 51.23%, better than mine (50% V-mAP). Since the model will be saved after each epoch, you could evaluate the model saved after 3, 4, and 5 epoch. On th other hand, I notice that your V-mAP@0.05 and V-mAP@0.1 results are significantly lower than mine, but I don't think it is a serious problem. V-mAP@0.05 and V-mAP@0.1 are low quality indicators. You just need to pay more attention to V-mAP@0.5.

[luyuanmeng]

Got it. Thank you.

…

---Original--- From: "Jianhua ***@***.***> Date: Mon, Nov 7, 2022 09:11 AM To: ***@***.***>; Cc: ***@***.******@***.***>; Subject: Re: [yjh0410/PyTorch_YOWO] accuracy not as good as YOWO-Plus's (Issue#6) @luyuanmeng Your ***@***.*** is 51.23%, better than mine (50% V-mAP). Since the model will be saved after each epoch, you could evaluate the model saved after 3, 4, and 5 epoch. On th other hand, I notice that your ***@***.*** and ***@***.*** results are significantly lower than mine, but I don't think it is a serious problem. ***@***.*** and ***@***.*** are low quality indicators. You just need to pay more attention to ***@***.*** — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>