Accurate 3D human pose estimation from single images is possible with sophisticated deep-net architectures that have been trained on very large datasets. However, this still leaves open the problem of capturing motions for which no such database exists. Manual annotation is tedious, slow, and error-prone. In this paper, we propose to replace most of the annotations by the use of multiple views, at training time only. Specifically, we train the system to predict the same pose in all views. Such a consistency constraint is necessary but not sufficient to predict accurate poses. We therefore complement it with a supervised loss aiming to predict the correct pose in a small set of labeled images, and with a regularization term that penalizes drift from initial predictions. Furthermore, we propose a method to estimate camera pose jointly with human pose, which lets us utilize multiview footage where calibration is difficult, e.g., for pan-tilt or moving handheld cameras. We demonstrate the effectiveness of our approach on established benchmarks, as well as on a new Ski dataset with rotating cameras and expert ski motion, for which annotations are truly hard to obtain.would be to annotate video data. However, achieving high accuracy would require a great deal of annotation, which is tedious, slow, and error-prone. As illustrated by Fig. 1, we therefore propose to replace most of the annotations by the use of multiple views, at training time only. Specifically, we use them to provide weak supervision and force the system to predict the same pose in all views.While such view consistency constraints increase accuracy, they are unfortunately not sufficient. For example, the network can learn to always predict the same pose, independently of the input image. To prevent this, we use a small set of images with ground-truth poses, which serve a dual purpose. First, they provide strong supervision during training. Second, they let us regularize the multi-view predictions by encouraging them to remain close to the predictions of a network trained with the scarce supervised data only.In addition, we propose to use a normalized pose distance to evaluate all losses involving poses. It disentangles pose from scale, and we found it to be key to maintain accuracy when the annotated data is scarce.Our experiments demonstrate the effectiveness of our weakly-supervised multi-view training strategy on several 1 arXiv:1803.04775v2 [cs.CV]

PURPOSE To compare biceps femoris long-head (BFlh) fascicle lengths (Lfs) obtained with different ultrasound-based approaches: 1) single ultrasound images and linear Lf extrapolation; 2) single ultrasound images and one of two different trigonometric equations (termed equations A and B); and 3) extended field of view (EFOV) ultrasound images. METHODS Thirty-seven elite alpine skiers (21.7±2.8 yrs) without a previous history of hamstring strain injury were tested. Single ultrasound images were collected with a 5 cm linear transducer from BFlh at 50% femur length and were compared with whole muscle scans acquired by EFOV ultrasound. RESULTS The intra-session reliability (ICC3,k = intraclass correlation coefficient) of Lf measurements was very high for both single ultrasound images (i.e., Lf estimated by linear extrapolation; ICC3,k = 0.96-0.99, SEM = 0.18 cm) and EFOV scans (ICC3,k = 0.91-0.98, SEM = 0.19 cm). Although extrapolation methods showed cases of overestimation and underestimation of Lf when compared with EFOV scans, mean Lf measured from EFOV scans (8.07±1.36 cm) was significantly shorter than Lf estimated by trigonometric equations A (9.98±2.12 cm, P<0.01) and B (8.57±1.59 cm, P=0.03), but not significantly different from Lf estimated with manual linear extrapolation (MLE) (8.40±1.68 cm, p=0.13). Bland-Altman analyses revealed mean differences in Lfs obtained from EFOV scans and those estimated from equation A, equation B and MLE of 1.91±2.1 cm, 0.50±1.0 cm and 0.33±1.0 cm, respectively. CONCLUSIONS The typical extrapolation methods used for estimating Lf from single ultrasound images are reliable within the same session, but not accurate for estimating BFlh Lf at rest with a 5-cm FOV. We recommend that EFOV scans are implemented to accurately determine intervention-related Lf changes in BFlh.

No abstract

Alpine ski racing is known to be a sport with a high risk of injury and a high proportion of time-loss injuries. In recent years, substantial research efforts with regard to injury epidemiology, injury etiology, potential prevention measures, and measures’ evaluation have been undertaken. Therefore, the aims of this review of the literature were (i) to provide a comprehensive overview of what is known about the aforementioned four steps of injury prevention research in the context of alpine ski racing; and (ii) to derive potential perspectives for future research. In total, 38 injury risk factors were previously reported in literature; however, a direct relation to injury risk was proven for only five factors: insufficient core strength/core strength imbalance, sex (depending on type of injury), high skill level, unfavorable genetic predisposition, and the combination of highly shaped, short and wide skis. Moreover, only one prevention measure (i.e. the combination of less-shaped and longer skis with reduced profile width) has demonstrated a positive impact on injury risk. Thus, current knowledge deficits are mainly related to verifying the evidence of widely discussed injury risk factors and assessing the effectiveness of reasonable prevention ideas. Nevertheless, the existing knowledge should be proactively communicated and systematically implemented by sport federations and sport practitioners.

BackgroundThere is limited knowledge about key injury risk factors in alpine ski racing, particularly for World Cup (WC) athletes.ObjectiveThis study was undertaken to compile and explore perceived intrinsic and extrinsic risk factors for severe injuries in WC alpine ski racing.MethodsQualitative study. Interviews were conducted with 61 expert stakeholders of the WC ski racing community. Experts’ statements were collected, paraphrased and loaded into a database with inductively derived risk factor categories (Risk Factor Analysis). At the end of the interviews, experts were asked to name those risk factors they believed to have a high potential impact on injury risk and to rank them according to their priority of impact (Risk Factor Rating).ResultsIn total, 32 perceived risk factors categories were derived from the interviews within the basic categories Athlete, Course, Equipment and Snow. Regarding their perceived impact on injury risk, the experts’ top five categories were: system ski, binding, plate and boot; changing snow conditions; physical aspects of the athletes; speed and course setting aspects and speed in general.ConclusionsSevere injuries in WC alpine ski racing can have various causes. This study compiled a list of perceived intrinsic and extrinsic risk factors and explored those factors with the highest believed impact on injury risk. Hence, by using more detailed hypotheses derived from this explorative study, further studies should verify the plausibility of these factors as true risk factors for severe injuries in WC alpine ski racing.