My custom dataset has mother_ Tiger and push_ Tiger has two categories, mother during reasoning_ Tiger did not successfully annotate.

[SCsleepy]

The model I used is dekr+hrnet, with a custom dataset training set of 90 images and a validation set of 10 images. Each image has four little tigers and one big tiger. I trained the weight inference of 500epoch (ap=0.31) and found that only the small tiger was labeled and the big tiger was not labeled. I trained the annotation image and inference result image as follows. I want to know what caused it.

[SCsleepy]

The dataset is in Coco format, and the following is an excerpt from the dataset
"images": [
{
"file_name": "1012_002251.jpg",
"height": 1440,
"width": 2560,
"id": 5
},
.....
"annotations": [
{
"segmentation": [
[
533.37,
131.01,
533.06,
128.98,
533.21,
127.73,
532.12,
126.33,
529.31,
126.01,
528.21,
127.57,
528.21,
129.29,
528.99,
130.23,
528.99,
130.7,
528.37,
132.1,
526.18,
134.29,
524.31,
136.79,
522.9,
139.29,
521.81,
142.41,
521.81,
144.6,
522.59,
145.22,
522.9,
145.69,
525.4,
143.66,
527.28,
142.41,
529.15,
140.69,
530.87,
138.35,
532.43,
137.1,
533.37,
135.85,
533.84,
133.67,
533.84,
132.1
],
[
524.78,
147.88,
524.31,
151.16,
524.31,
154.28,
524.15,
158.03,
524.47,
161,
524.93,
164.28,
525.4,
168.34,
525.25,
170.83,
524.47,
172.86,
522.75,
173.8,
522.59,
174.43,
529.15,
174.74,
529.46,
173.8,
531.18,
172.4,
531.81,
171.62,
532.12,
168.34,
531.18,
155.84,
531.34,
149.6,
531.96,
146,
531.49,
145.85,
527.59,
148.35,
526.5,
147.88,
525.25,
147.25
]
],
"keypoints": [
1425.302987,
141.4731466,
2,
0,
0,
0,
1401.679555,
169.1721407,
2,
1316.746787,
135.1248356,
2,
1276.117565,
183.3720357,
2,
0,
0,
0,
1378.960281,
213.20912,
2,
1207.555755,
137.664162,
2,
1307.859145,
216.3832779,
2,
0,
0,
0,
0,
0,
0,
1316.746787,
87.51246713,
2,
1246.915313,
65.29336182,
2,
1168.831028,
103.3832566,
2,
1152.325407,
134.4900041,
2,
1241.201829,
203.0518147,
2
],
"num_keypoints": 12,
"area": 41244.15965392148,
"iscrowd": 0,
"image_id": 5,
"bbox": [
1152.325407,
65.29336182,
1425.302987,
216.3832779
],
"category_id": 2,
"id": 26
},
{
"segmentation": [
[
533.37,
131.01,
533.06,
128.98,
533.21,
127.73,
532.12,
126.33,
529.31,
126.01,
528.21,
127.57,
528.21,
129.29,
528.99,
130.23,
528.99,
130.7,
528.37,
132.1,
526.18,
134.29,
524.31,
136.79,
522.9,
139.29,
521.81,
142.41,
521.81,
144.6,
522.59,
145.22,
522.9,
145.69,
525.4,
143.66,
527.28,
142.41,
529.15,
140.69,
530.87,
138.35,
532.43,
137.1,
533.37,
135.85,
533.84,
133.67,
533.84,
132.1
],
[
524.78,
147.88,
524.31,
151.16,
524.31,
154.28,
524.15,
158.03,
524.47,
161,
524.93,
164.28,
525.4,
168.34,
525.25,
170.83,
524.47,
172.86,
522.75,
173.8,
522.59,
174.43,
529.15,
174.74,
529.46,
173.8,
531.18,
172.4,
531.81,
171.62,
532.12,
168.34,
531.18,
155.84,
531.34,
149.6,
531.96,
146,
531.49,
145.85,
527.59,
148.35,
526.5,
147.88,
525.25,
147.25
]
],
"keypoints": [
1811.28051,
220.827099,
2,
1863.971531,
312.8776781,
2,
1787.15691,
295.1023938,
2,
0,
0,
0,
0,
0,
0,
1895.71311,
262.7259832,
2,
1879.842321,
325.5743097,
2,
1800.488373,
140.8383199,
2,
1732.561394,
189.0855199,
2,
0,
0,
0,
1771.920952,
185.2765305,
2,
1859.52771,
217.0181095,
2,
1863.971531,
178.2933831,
2,
1862.067036,
124.3326988,
2,
0,
0,
0,
1806.836689,
111.0012356,
2
],
"num_keypoints": 12,
"area": 35007.96524681013,
"iscrowd": 0,
"image_id": 5,
"bbox": [
1732.561394,
111.0012356,
1895.71311,
325.5743097
],
"category_id": 2,
"id": 27
},
{
"segmentation": [
[
533.37,
131.01,
533.06,
128.98,
533.21,
127.73,
532.12,
126.33,
529.31,
126.01,
528.21,
127.57,
528.21,
129.29,
528.99,
130.23,
528.99,
130.7,
528.37,
132.1,
526.18,
134.29,
524.31,
136.79,
522.9,
139.29,
521.81,
142.41,
521.81,
144.6,
522.59,
145.22,
522.9,
145.69,
525.4,
143.66,
527.28,
142.41,
529.15,
140.69,
530.87,
138.35,
532.43,
137.1,
533.37,
135.85,
533.84,
133.67,
533.84,
132.1
],
[
524.78,
147.88,
524.31,
151.16,
524.31,
154.28,
524.15,
158.03,
524.47,
161,
524.93,
164.28,
525.4,
168.34,
525.25,
170.83,
524.47,
172.86,
522.75,
173.8,
522.59,
174.43,
529.15,
174.74,
529.46,
173.8,
531.18,
172.4,
531.81,
171.62,
532.12,
168.34,
531.18,
155.84,
531.34,
149.6,
531.96,
146,
531.49,
145.85,
527.59,
148.35,
526.5,
147.88,
525.25,
147.25
]
],
"keypoints": [
1717.104888,
1040.819085,
2,
1623.149814,
1046.532569,
2,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1608.548688,
1153.184274,
2,
1576.172277,
1098.588758,
2,
0,
0,
0,
0,
0,
0,
1609.183519,
1365.218022,
2,
1587.599246,
1356.965212,
2,
1663.144204,
1139.852811,
2,
1653.62173,
1242.060696,
2,
1658.065551,
1364.583191,
2,
1649.177909,
1407.751738,
2,
0,
0,
0
],
"num_keypoints": 10,
"area": 51712.77684844703,
"iscrowd": 0,
"image_id": 5,
"bbox": [
1576.172277,
1040.819085,
1717.104888,
1407.751738
],
"category_id": 2,
"id": 28
},
{
"segmentation": [
[
533.37,
131.01,
533.06,
128.98,
533.21,
127.73,
532.12,
126.33,
529.31,
126.01,
528.21,
127.57,
528.21,
129.29,
528.99,
130.23,
528.99,
130.7,
528.37,
132.1,
526.18,
134.29,
524.31,
136.79,
522.9,
139.29,
521.81,
142.41,
521.81,
144.6,
522.59,
145.22,
522.9,
145.69,
525.4,
143.66,
527.28,
142.41,
529.15,
140.69,
530.87,
138.35,
532.43,
137.1,
533.37,
135.85,
533.84,
133.67,
533.84,
132.1
],
[
524.78,
147.88,
524.31,
151.16,
524.31,
154.28,
524.15,
158.03,
524.47,
161,
524.93,
164.28,
525.4,
168.34,
525.25,
170.83,
524.47,
172.86,
522.75,
173.8,
522.59,
174.43,
529.15,
174.74,
529.46,
173.8,
531.18,
172.4,
531.81,
171.62,
532.12,
168.34,
531.18,
155.84,
531.34,
149.6,
531.96,
146,
531.49,
145.85,
527.59,
148.35,
526.5,
147.88,
525.25,
147.25
]
],
"keypoints": [
2067.290874,
1336.436316,
2,
2184.890178,
1339.844945,
2,
2076.756227,
1437.407016,
2,
1937.201772,
1272.619467,
2,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1780.31929,
889.190065,
2,
1522.229628,
1011.693453,
2,
1780.31929,
987.9063873,
2,
0,
0,
0,
2102.634029,
1210.315451,
2,
2061.006664,
999.7999201,
2,
1970.615815,
822.5862813,
2,
1854.059193,
746.4676713,
2,
1720.851626,
526.4373144,
2
],
"num_keypoints": 12,
"area": 603663.683495592,
"iscrowd": 0,
"image_id": 5,
"bbox": [
1522.229628,
526.4373144,
2184.890178,
1437.407016
],
"category_id": 1,
"id": 30
},
....
"categories": [
{
"id": 1,
"name": "mother_tiger"
},
{
"id": 2,
"name": "pup_tiger"
}

[Ben-Louis]

Hi, thanks for using MMPose. It seems like there may be a misdetection issue. There are a couple of possible reasons for this:

1. The [rescore net](the https://github.com/open-mmlab/mmpose/blob/537bd8e543ab463fb55120d5caaa1ae22d6aaf06/configs/body_2d_keypoint/dekr/coco/dekr_hrnet-w32_8xb10-140e_coco-512x512.py#L101C6-L101C6) from DEKR may not be suitable for your dataset, and should be removed.
2. If your training set is too small, it may not be able to effectively train a large network like HRNet. Consider training DEKR on a larger dataset like AP10k, and then fine-tuning it with your own data.